ES2759513T3 - Entropic Encoding and Decoding Scheme - Google Patents

Abstract

Entropic coding apparatus comprising a decomposer (136) configured to convert a sequence (138) of syntax elements having a range of values that is subdivided into a sequence of N partitions (1401-3) into a sequence (106) of source symbols (106) individually decomposing at least a subset of the syntax elements into a respective number n of source symbols if with i = 1 ... n, the respective number n of source symbols depending on which of the sequences in N partitions (1401-3) drops a value z of the respective syntax elements, so that a sum of values of the respective number of source symbols if it produces z, and, if n> 1, for all i = 1 ... n -1, the value of if corresponds to a rank of the i-th partition; a subdivider (100) configured to subdivide the source symbol sequence (106) into a first source symbol subsequence (108) and a second source symbol subsequence (110) such that all source symbols sx, x being a member of a first subset of {1 ... N}, are contained within the first subsequence (108) and all source symbols sy, where y is a member of a second subset of {1 ... N} that is disjoint with respect to first subset, are contained within the second subsequence (110); a VLC encoder (102) configured to encode in a symbol level manner the source symbols of the first subsequence (108); and an arithmetic encoder (104) configured to encode the second subsequence (110) of source symbols, characterized in that the subgroup of the syntax elements encompasses absolute transform coefficient levels of absolute transform coefficients of image transform blocks with the Absolute transform coefficient levels of a respective transform block representing a prediction residual and arranged within the sequence (138) of syntax elements according to a scanning path leading through the absolute transform coefficients of the respective transform blocks, where the decomposer is configured to adapt one or more of the boundaries between the divisions during the decomposition of the absolute transform coefficient levels of the absolute transform coefficients of a respective transform block depending on coefficient levels of absolute transformation already codi Absolute transform coefficient plots of the respective preceding transform blocks in the scanning order or depending on a position of the absolute transform coefficient level to be currently decomposed in the scanning order, or based on an evaluation of the coefficient levels of Absolute transform already reconstructed from close transform coefficients, either spatially or in scanning order, to the position of the absolute transform coefficient level to be decomposed currently.

Description

DESCRIPTION

Entropic Encoding and Decoding Scheme

The present invention relates to entropic encoding and decoding and can be used in applications such as, for example, video and audio compression.

Entropic encoding, in general, can be considered as the most generic form of lossless data compression. Lossless compression is intended to represent discrete data with fewer bits than is required to represent the original data but without any loss of information. Discrete data can be provided in the form of text, graphics, images, video, audio, voice, facsimile, medical data, weather data, financial data, or any other form of digital data.

In entropic encoding, the high-level characteristics specific to the underlying discrete data source are often neglected. Consequently, any data source is considered to be provided as a sequence of font symbols taking values in a given m-ary alphabet and characterized by a corresponding (discrete) probability distribution {p- ..., pm}. In these abstract settings, the lower bounding of any entropic encoding method in terms of expected codeword length in bits per symbol is provided by entropy

- Z P ⁱ lo ^ ² P ⁱ

H = ⁱ = ¹ (A1) Huffman codes and arithmetic codes are well-known examples of practical codes capable of approaching the entropy limit (in a certain sense). For a fixed probability distribution, Huffman codes are relatively easy to construct. The most attractive property of Huffman codes is that they can be implemented efficiently using variable length code tables (VLC). However, when working with statistics of variable origin over time, that is, probabilities of changing symbols, the adaptation of the Huffman code and its corresponding VLC tables is quite demanding, both in terms of algorithmic complexity and in terms of costs. of implementation. Also, in the case of having a dominant alphabet value with pk > 0.5, the redundancy of the corresponding Huffman code (without the use of any alphabet extension such as string length encoding) can be quite substantial. Other drawbacks to Huffman codes come from the fact that in dealing with higher order probability modeling, multiple sets of VLC tables may be required. Arithmetic coding, on the other hand, while substantially more complex than VLC, offers the advantage of more consistent and appropriate handling when dealing with adaptive and higher-order probability modeling as well as highly skewed probability distributions. . Actually, this feature is basically the result of the fact that arithmetic coding provides a mechanism, at least conceptually, to map any given estimated probability value in a more or less direct way to a part of the resulting codeword. Providing itself with such an interface, arithmetic coding allows a clean separation between probability modeling and probability estimation tasks, on the one hand, and actual entropic coding, that is, mapping from symbols to codewords, by other side.

An alternative to arithmetic coding and VLC coding is PIPE coding. To be more precise, in PIPE coding, the unit interval is partitioned into a small set of discrete probability intervals to channel the coding process along with the probability estimates of random symbol variables. According to this partitioning, an input sequence of discrete source symbols with arbitrary alphabet sizes can be mapped to a sequence of alphabet symbols, and each of the alphabet symbols is assigned to a particular probability range that is encoded, its time, through a specially dedicated entropic coding process. With each interval represented by a fixed probability, the probability interval partitioning entropy (PIPE) encoding process can be based on the design and application of simple variable-to-variable length codes. Probability modeling can be either fixed or adaptive. However, although PIPE encoding is significantly less complex than arithmetic encoding, it still has a higher complexity than VLC encoding.

Therefore, it would be favorable to have an entropic encoding scheme available that allows a better compromise between the encoding complexity on the one hand and the compression efficiency on the other hand, even when compared to PIPE encoding which already combines the advantages of both arithmetic coding as well as VLC coding.

In addition, in general, it would be favorable to have an entropic coding scheme available that allows achieving a better compression efficiency per se, at a moderate coding complexity.

WO 2008/129021 A2 relates to scalable compression of time-consistent 3D mesh sequences. Regarding the quantification and entropic coding, the document describes that the forecast errors of the mesh vectors are compressed component by component. In particular, the components are mapped to the number of integers, that is, signed, and a value. If the value is greater than imax, the value is decomposed into two parts, namely imax and imax-value. imax and the value, as long as it is less than imax, is encoded by arithmetic encoding. The imax-quantity is encoded by Golomb encoding.

Thus, it is an object of the present invention to provide an entropic coding concept that satisfies the previously identified demand, that is, allows to achieve a better compromise between the coding complexity on the one hand and the compression efficiency on the other hand.

This object is achieved by an apparatus according to claim 1 or 3 and a method according to claim 5 or 6. The present invention is based on the idea of decomposing a range of values of the respective syntax elements into a sequence of n encoded partitions of the components of the z syntax element values that fall within the respective partitions separately with at least one by VLC encoding and at least one by arithmetic encoding.

Accordingly, in accordance with the embodiments of the present invention, the syntax elements are decomposed into a respective number of n source symbols if, where i = 1 ... n, the respective number n of source symbols depending on which of a sequence of n partitions (140i -3) into which a range of values of the respective syntax elements is subdivided, falls within a z-value of the respective syntax elements, such that a sum of values of the respective number of source symbols if it produces z, and, if n> 1, for all i = 1.n-1, the value of si corresponds to an interval of the i-th partition.

Preferred aspects of the present invention are the subject of the appended dependent claims.

Preferred embodiments of the present invention are described below with respect to the figures. These embodiments represent, insofar as their subject matter is not covered by the claims, only useful examples to understand the invention. Among the figures,

Fig. 1a shows a block diagram of an entropic coding apparatus according to one embodiment;

Fig. 1b shows a schematic diagram illustrating a possible decomposition of syntax elements into source symbols according to one embodiment;

Fig. 1c shows a flow diagram illustrating a possible mode of operation of the decomposer of Fig. 1a in decomposing syntax elements into source symbols according to one embodiment;

Fig. 2a shows a block diagram of an entropic decoding apparatus according to one embodiment;

Fig. 2b shows a flow chart illustrating a possible mode of operation of the composer of Fig. 2a in composing syntax elements from source symbols according to one embodiment;

Fig. 3 shows a block diagram of a PIPE encoder according to an embodiment that can be used in Fig. 1;

Fig. 4 shows a block diagram of a PIPE decoder suitable for decoding a bit stream generated by the PIPE encoder of Fig. 3, according to an embodiment that can be used in Fig. 2;

Fig. 5 shows a schematic diagram illustrating a data packet with multiplexed partial bit streams according to one embodiment;

Fig. 6 shows a schematic diagram illustrating a data packet with alternative segmentation using fixed-size segments according to a further embodiment;

Fig. 7 shows a block diagram of a PIPE encoder using partial bitstream interleaving according to one embodiment;

Fig. 8 shows schematic illustration examples for the status of a codeword buffer on the encoder side of Fig. 7 according to one embodiment;

Fig. 9 shows a block diagram of a PIPE decoder using partial bitstream interleaving according to one embodiment;

Fig. 10 shows a block diagram of a PIPE decoder according to an embodiment that uses codeword interleaving using a simple set of codewords;

Fig. 11 shows a block diagram of a PIPE encoder according to an embodiment using fixed-length bitstream interleaving;

Fig. 12 shows a schematic illustrating examples for the status of a global bit buffer on the encoder side of Fig. 11 according to one embodiment;

Fig. 13 shows a block diagram of a PIPE decoder according to an embodiment using fixed length bitstream interleaving;

Fig. 14 shows a graph for the illustration of an optimal probability interval outlined in K =

4 intervals assuming a uniform probability distribution at (0, 0.5];

Fig. 15 shows a schematic diagram illustrating a binary event tree for an LPB probability of P = 0.38 and an associated variable length code obtained by the Huffman algorithm;

Fig. 16 shows a graph from which the relative increase in bit rate p (p,

C) for optimal C codes given a maximum number of table entries Lm;

Fig. 17 shows a graph illustrating the rate increase for partitioning the theoretically optimal probability interval into K = 12 intervals and a real design with V2V codes with a maximum number of table entries Lm = 65;

Fig. 18 shows a schematic diagram illustrating an example for converting a ternary choice tree to a full binary choice tree;

Fig. 19 shows a block diagram of a system comprising an encoder (left side) and decoder (right side) according to one embodiment;

Fig. 20 shows a block diagram of an entropic coding apparatus according to one embodiment;

Fig. 21 shows a block diagram of an entropic decoding apparatus according to one embodiment;

Fig. 22 shows a block diagram of an entropic coding apparatus according to a further embodiment;

Fig. 23 shows a schematic diagram illustrating examples of the state of a global bit buffer on the encoder side of Fig. 22 according to one embodiment;

Fig. 24 shows a block diagram of an entropic decoding apparatus according to a further embodiment of the present application.

Before various embodiments of the present application are described below with respect to the figures, it is noted that the same reference signs are used throughout all the figures to indicate the same or equivalent elements in these figures, and the description of these Elements presented in any of the previous figures can also be applied to any of the following figures as long as the previous description does not conflict with the description of the current figures.

Fig. 1a shows an entropic coding apparatus according to an embodiment of the present application. The apparatus comprises a subdivider 100, a VLC 102 encoder and a PIPE 104 encoder.

Subdivider 100 is configured to subdivide a source symbol sequence 106 into a first source symbol subsequence 108 and a second source symbol subsequence 110. The VLC 102 Encoder it has an input thereof connected to a first output of subdivider 100 and is configured to convert in a symbol level manner, the source symbols of the first subsequence 108 into code words that form a first bit stream 112. The VLC encoder 102 may comprise a lookup table and individually use the source symbols as an index for searching, by source symbol, for a respective code word in the lookup table. The VLC encoder produces the last codeword, and proceeds to the next source symbol in subsequence 110 to produce a sequence of codewords in which each codeword is associated with exactly one of the source symbols within subsequence 110 Codewords can be of different lengths and can be defined so that no codeword prefixes any of the other codewords. Additionally, the lookup table can be static.

The PIPE encoder 104 has an input thereof connected to a second output of the subdivider 100 and is configured to encode the second subsequence 110 of source symbols, represented in the form of an alphabet symbol sequence, and comprises an allocator 114 configured to assign a measure for an estimate of a probability distribution among possible values that the respective alphabet symbols can assume, for each alphabet symbol in the sequence of alphabet symbols based on the information contained within the previous alphabet symbols in the sequence of alphabet symbols, a plurality of entropic encoders 116 each of which is configured to convert the alphabet symbols directed to the respective entropic encoder into a respective second bitstream 118, and a selector 120 configured to direct each alphabet symbol of the second subsequence 110 to one selected from the plurality of codi entropic factors 116, the selection of the aforementioned measure depending on the estimation of the probability distribution assigned to the respective alphabet symbol. The association between source symbols and alphabet symbols may be such that each alphabet symbol is associated with only exactly one source symbol of the subsequence 110 to be represented, possibly together with additional alphabet symbols from the sequence of alphabet symbols that can be followed immediately. each other, this source symbol.

As described in more detail below, source symbol sequence 106 may be a bitstream syntax element sequence that can be parsed. The bitstream that can be parsed can, for example, represent video and / or audio content in a scalable or non-scalable form by representing the syntax elements, for example, transform coefficient levels, motion vectors, reference indices of motion picture, scale factors, energy values of the audio envelope or the like. The syntax elements may, in particular, be of a different type or category with the syntax elements having the same type, for example, the same meaning within the bitstream that can be parsed but with respect to different parts of it, such as different images, different macroblocks, different spectral components, or the like, while different type syntax elements may have a different meaning within the bit stream, such as a motion vector has a different meaning than a syntax element representing a level transform coefficient representing the prediction of residual motion.

Subdivider 100 may be configured to subdivide depending on the type of syntax elements. That is, subdivider 100 can direct syntax elements from a first group of types to the first subsequence 108 and direct syntax elements from a second group of types other than the first group, to the second subsequence 110. The subdivision made by the subdivider 100 can be designed such that the symbol statistics of the syntax elements within the subsequence 108 is suitable to be encoded by VLC by encoder VLC 102, i.e. results in almost as little entropy as possible despite the use of encoding VLC and its restriction regarding its suitability for certain symbol statistics as described in the introductory part of the memory of the present application. On the other hand, subdivider 100 can direct all other syntax elements in second subsequence 110 so that these syntax elements that have symbol statistics that are not suitable for VLC encoding, are encoded by a more complex PIPE 104 encoder , but more efficient - in terms of the compression ratio.

As is also the case with the embodiments described in greater detail in the following figures, the PIPE encoder 104 may comprise a symbolizer 122 configured to individually map each syntax element of the second subsequence 110 into a respective partial sequence of alphabet symbols, together with the formation of the above-mentioned sequence 124 of alphabet symbols. In other words, symbolizer 122 may not be present if, for example, the subsequence 110 source symbol is already represented as respective partial sequences of alphabet symbols. Symbolizer 122 is, for example, advantageous if the source symbols within subsequence 110 are of different alphabets, and especially, alphabets having different numbers of possible alphabet symbols. Specifically, in this case, symbolizer 122 is capable of harmonizing the alphabets of symbols arriving within subflow 110. Symbolizer 122 can, for example, be performed as a binary converter configured to convert symbols arriving within binary to binary. the subsequence 110.

As mentioned above, syntax elements can be of a different type. This can also be true for syntax elements within subflow 110. Symbolizer 122 can be configured then to perform individual mapping of the subsequence 110 syntax elements using a symbolization mapping scheme, such as a different binary conversion scheme for different type syntax elements. Examples for specific binary conversion schemes are presented in the following description, such as a unary binary conversion scheme, an order 0 or order 1 exp-Golomb binary conversion scheme, for example, or a conversion scheme to truncated unary binary, a 0-order exp-Golomb rearranged and truncated binary conversion scheme, or an unsystematic binary conversion scheme.

Accordingly, entropic encoders 116 could be configured to operate on a binary alphabet. Finally, it should be noted that symbolizer 122 can be considered as part of the same PIPE encoder 104 as shown in Fig. 1a. As an alternative, however, the binary converter can be considered external to the PIPE encoder.

Similar to the last note, it should be noted that allocator 114, while shown connected in series between symbolizer 122 and selector 120, can alternatively be viewed as connected between an output of symbolizer 124 and a first input of selector 120, an output of allocator 114 being connected to another input of selector 120 as described below with respect to FIG. 3. Indeed, allocator 114 accompanies each alphabet symbol with the aforementioned measurement for an estimate of the probability distribution.

As regards the output of the entropic encoding apparatus of Fig. 1a, it is composed of the first bitstream 112 produced by the VLC encoder 102 and the plurality of second bitstreams 118 produced by the plurality of encoders Entropics 116. As described further below, all of these bit streams can be transmitted in parallel. Alternatively, they can be interleaved into a common bitstream 126 through the use of an interleaver 128. The embodiments of Figs. 22-24 show examples of such bitstream interleaving. As further shown in FIG. 1, the same PIPE encoder 104 may include its own interleaver 130 to interleave the plurality of second bit streams 118 into a common PIPE encoded bit stream 132. The possibilities for said interleaver 130 can be deduced from the following embodiments described with respect to Figs. 5 to 13. Bit stream 132 and bit stream 112 may, in a parallel configuration, represent the output of the entropic encoding apparatus of Fig. 1a. Alternatively, another interleaver 134 may interleave both bitstreams in which case interleaver 130 and 134 would form two stages of a two stage interleaver 128.

As described above, subdivider 100 can perform subdivision at the level of syntax elements, that is, the source symbols on which subdivider 100 operates can be complete syntax elements, or in other words, subdivider 100 can operate on syntax element units.

However, in accordance with an alternate embodiment, the entropic encoding apparatus of FIG. 1a may include decomposer 136 to decompose syntax elements into bitstream 138 that can be individually parsed into one or more source symbols in the sequence. source symbol 106 entering subdivider 100. In particular, decomposer 136 can be configured to convert syntax element sequence 138 to source symbol sequence 106 by individually decomposing each syntax element into a respective integer number of source symbols. The integer can vary between syntax elements. In particular, some of the syntax elements may even be left unchanged by decomposer 136, while other syntax elements are decomposed into exactly two, or at least two, source symbols. Subdivider 100 can be configured to direct one of the source symbols of said decomposed syntax elements to the first subsequence 108 of the source symbols and another of the source symbols of the same decomposed syntax element to the second subsequence 110 of source symbols. As mentioned above, the syntax elements within bit stream 138 can be of a different type, and decomposer 136 can be configured to perform individual decomposition depending on the type of the syntax element. The decomposer 136 preferably performs the individual decomposition of the syntax elements so that there is a predetermined unique inverse mapping used later on the decoding side, from the integer number of source symbols to the respective syntax element, common to all elements of syntax.

For example, decomposer 136 can be configured to decompose z syntax elements in parsable stream 138, into two source symbols x and y so that z = xy, z = x - y, z = x ■ yoz = x: and. With this measure, the subdivider 100 can decompose the syntax elements into two components, namely source symbols of the source symbol stream 106, one of which is suitable to be encoded by VLC in terms of compression efficiency, such as x, and the another of which is not suitable for VLC encoding and is therefore passed to the second subflow 110 instead of the first subflow 108, such as and. The decomposition used by decomposer 136 need not be bijective. However, as mentioned above there would be an inverse mapping that would allow a unique recovery of the syntax elements from possible decompositions among which decomposer 136 can choose if the decomposition is not bijective.

So far, different possibilities for handling different syntax elements have been described. If such syntax or case elements exist, it is optional. The embodiments described hereinafter, however, focus on syntax elements that are decomposed by decomposer 136 according to the following principle.

As shown in Fig. 1b, decomposer 136 is configured to decompose certain z-syntax elements into bit stream 138 that can be parsed in stages. There may be two or more stages. The steps are for dividing the z-syntax element value range into two or more adjacent sub-ranges or sub-ranges as shown in Fig. 1c. The syntax element value range can have two infinite endpoints, just one or it can have defined endpoints. In Fig. 1c, the range of syntax element values is subdivided exemplary into three 1401-3 partitions. As shown in Fig. 1b, if the syntax element is greater than or equal to the delimitation 142 of the first partition 1401, ie the upper limit that separates partitions 1401 and 1402, then the syntax element is subtracted delimitation limit1 of the first partition 1403 and z is checked again to see if it is even greater than or equal to the delimitation 144 of the second partition 1402, that is, the upper limit that separates partitions 1402 and 1403. If z ' is greater than or equal to the bounding 144, then z 'subtracts the bounding bound2 from the second partition 1402 resulting in z' '. In the first case where z is smaller than limit1, the z syntax element is sent to subdivider 100 clearly, in case z is between limit1 and limit2, the z syntax element is sent to subdivider 100 in a tuple (limit1, z ') where z = limit1 + z', and in case z is above limit2, the syntax element z is sent to subdivider 100 as a triplet (limit1, limit2-limit1, z ') being z = limit1 + limit2 z '. The first (or only) component, that is to say zo limit1, forms a first source symbol to be encoded by subdivisor 100, the second component, that is to say z 'or limit2-limit1, forms a second source symbol to be encoded by subdivider 100, if present, and the third component, i.e. z '', forms a third source symbol to be encoded by subdivider 100, if present. Therefore, according to the embodiment of Fig. 1b and 1c, the syntax element maps to any of 1 to 3 source symbols, but the generalizations about a smaller or larger maximum number of source symbols is easily deducible from the description above, and such alternatives will also be described below.

In any case, all of these different different components or source symbols are in accordance with the embodiments below, encoded with encoding among the alternatives. At least one of them is forwarded by the subdivider to the PIPE encoder 104, and at least one of them is sent to the VLC 102 encoder.

Particular advantageous embodiments are described in more detail below.

Having previously described an embodiment for an entropic encoding apparatus, an embodiment for an entropic decoding apparatus is described with respect to FIG. 2a. The entropic decoding apparatus of FIG. 2a comprises a VLC 200 decoder and a PIPE 202 decoder. The VLC 200 decoder is configured to reconstruct in a code level manner the source symbols of a first codeword subsequence 204 of a first bitstream 206. The first bitstream 206 is equal to bitstream 112 of FIG. 1, and the same applies to subsequence 204 as it relates to subsequence 108 of FIG. 1a. The PIPE decoder 202 is configured to reconstruct a second subsequence 208 of source symbols, represented in the form of an alphabet symbol sequence, and comprises a plurality of entropic decoders 210, an allocator 212, and a selector 214. The plurality of entropic decoders 210 they are configured to convert a respective one of the second bitstreams 216 into alphabet symbols of the alphabet symbol sequence. The allocator 212 is configured to assign a measure of an estimate of a probability distribution among possible values that the respective alphabet symbols can assume, to each alphabet symbol in the sequence of alphabet symbols representing the second subsequence 208 of source symbols to be reconstructed, based on the information contained within the previously reconstructed alphabet symbols of the alphabet symbol sequence. To this end, allocator 212 may be connected in series between an output of selector 214 and an input thereof, while additional inputs of selector 214 have outputs from entropy decoders 210 respectively connected thereto. Selector 214 is configured to retrieve each alphabet symbol in the sequence of alphabet symbols from one selected from the plurality of entropic decoders 210, the selection of measurement assigned to the respective alphabet symbol depending. In other words, selector 214 together with allocator 212 is operative to retrieve the alphabet symbols obtained by the entropic decoders 210 in an order among the entropic decoders 210 obtained by studying the information contained within previous alphabet symbols of the alphabet symbols sequence. In other words, mapper 212 and selector 214 are able to reconstruct the original order of the alphabet symbols from alphabet symbol to alphabet symbol. In conjunction with predicting the next alphabet symbol, the allocator 212 is able to determine the aforementioned measure of the probability distribution estimate for the respective alphabet symbol by using which selector 214 selects among entropic decoders 210 to retrieve the actual value of this alphabet symbol. To be even more precise, and as will be described in more detail below, PIPE decoder 202 can be configured to reconstruct symbol sequence 208 font, represented in the form of the alphabet symbol sequence, in response to requests for alphabet symbols that sequentially request the alphabet symbols, and allocator 212 can be configured to assign each request for an alphabet symbol in the sequence of alphabet symbols representing the second subsequence (208) of source symbols to be reconstructed, the aforementioned measure of an estimate of a probability distribution among the possible values that the respective alphabet symbols can assume. Accordingly, selector 214 can be configured to retrieve, for each request for an alphabet symbol from the alphabet symbol sequence representing the second subsequence (208) of source symbols to be reconstructed, the respective alphabet symbol from the symbol sequence of alphabet from one selected from the plurality of entropic decoders 210, the selection of the size assigned to the respective request depending on the respective alphabet symbol. The match between requests on the decoding side on the one hand, and data flow or encoding on the coding side on the other hand, will be described in more detail with respect to Fig. 4.

Since the first source symbol subsequence 214 and the second source symbol subsequence 208 commonly form a common source symbol sequence 210, the entropic decoding apparatus of FIG. 2a may optionally comprise a recombiner 220 configured to recombine the first subsequence 204 and the second subsequence 208 to obtain the common sequence 218 of source symbols. This common sequence 208 of source symbols produces a reconstruction of sequence 106 of FIG. 1a.

In accordance with the description presented above with respect to Fig. 1, the source symbols of the first and second subsequences 204 and 208 may be syntax elements of a bitstream that can be parsed. In this case, the recombiner 220 could be configured to reconstruct this bitstream that can be parsed from sequence 218 of the syntax elements by interleaving the source symbols arriving through the first and second subsequences 204 and 208 in an order prescribed by some parsing rule that defines an order between the syntax elements. In particular, the syntax elements can be, as described above, of a different type and the recombiner 220 can be configured to retrieve or request syntax elements of a first group of types from the VLC decoder 200 through subflow 204, and syntax elements of a second type from the PIPE decoder 202 through subflow 208. Accordingly, as long as the parser rule indicates that a syntax element of a type within the first group is inline next, the recombiner 202 inserts a current source symbol from subsequence 204 into common sequence 218 and otherwise from subsequence 208.

In the same way, the PIPE decoder 202 could comprise a de-symbolizer 222 connected between the output of the selector 214 and an input of the recombiner 220. Similarly to the previous description with respect to Fig. 1, the de-symbolizer 222 could be considered as external to the PIPE decoder 202 and could even be arranged behind the recombiner 202, ie on the output side of the recombiner 220, as an alternative. The de-symbolizer 222 could be configured to remap, in units of partial alphabet symbol sequences, the alphabet symbol sequence 224 produced by selector 214 into the source symbols, ie, subsequence 208 syntax elements. Similarly to recombiner 220, deimbolizer 222 is aware of the construction of possible partial sequences of alphabet symbols. In particular, the de-symbolizer 222 can analyze recently received alphabet symbols from selector 214 to determine as to whether these recently received alphabet symbols produce a valid partial sequence of alphabet symbols associated with a respective value of the respective syntax element, or if this is not the case, and what alphabet symbol is missing next. Even in other words, symbolizer 222 is aware, at any time, whether additional alphabet symbols are to be received from selector 214 to end reception of a respective syntax element or not, and consequently, to which syntax element it belongs. a respective one of the alphabet symbols produced by selector 214. To this end, the de-symbolizer 222 can use a symbolization (de) mapping scheme that differentiates syntax elements of different type. Similarly, mapper 212 is aware of the association of a current alphabet symbol to be retrieved from any of the entropic decoders 210 by selector 214, to a respective one of the syntax elements, and can set the aforementioned measure of estimation of a probability distribution of this alphabet symbol accordingly, i.e. it depends on the type of associated syntax element. Still further, allocator 212 can differentiate between different alphabet symbols belonging to the same partial sequence of a current alphabet symbol and can set the measure of the probability distribution estimate differently for these alphabet symbols. Details in this regard are described in more detail below. As described herein, mapper 212 can be configured to assign contexts to alphabet symbols. The assignment may depend on the type of syntax element and / or the position within the partial sequence of the alphabet symbols of the current syntax element. As soon as allocator 212 has assigned a context to the current alphabet symbol to be retrieved from any of the entropy decoders 210 by selector 214, the alphabet symbol may intrinsically have the mean of the estimate of the probability distribution associated with it. since each context has its estimation measure associated with it. Additionally, the context - and its associated probability distribution estimation measure - can be adapted according to the current statistics of the respective context alphabet symbols that have been retrieved from the entropic decoders 210 thus far. I know They present in more detail below details in this regard.

Similar to the explanation above in Fig. 1, it may be possible that the correspondence between the aforementioned source symbols of subsequences 204 and 208 in syntax elements is not a one-to-one correspondence. Rather, the syntax elements may have been decomposed into an integer number of source symbols by varying the number, eventually, between the syntax elements but in any case being greater than one at least for one syntax element. As noted above, the embodiments below focus on handling this class of syntax elements, and syntax elements from other classes may not even be present.

For handling the aforementioned syntax elements, the entropic decoding apparatus of Fig. 2a may comprise a composer 224 configured to redo the decomposition performed by the decomposer 136 of Fig. 1a. In particular, the composer 224 can be configured to compose the sequence 226 of syntax elements from the source symbols of the sequence 218 or, if the recombiner 220 is absent, the subsequences 204 and 208, composing each syntax element individually from a respective integer number of source symbols one of the source symbols of the integer number of source symbols belonging to the first subsequence 204 and another of the source symbols belonging to the integer number of source symbols of the same syntax element to the second subsequence 208. With this measure , certain syntax elements may have been decomposed on the encoder side to separate components suitable for VLC decoding from a remaining component to be passed through a PIPE decoding path. Similar to the explanation above, the syntax element can be a different type and composer 224 can be configured to perform individual composition depending on the type of syntax elements. In particular, the composer 224 can be configured to obtain the respective syntax elements by logical or mathematical combination of the integer number of source symbols of the respective syntax element. For example, Composer 224 can be configured to apply, to each syntax element, -. : or ■ to the first and second source symbols of a syntax element.

As described above, the embodiments described herein below, however, focus on syntax elements that are decomposed by decomposer 136 according to Figs. 1b and 1c and the alternatives described in relation to them. Fig. 2a shows how composer 224 can work to reconstruct these syntax elements from their source symbols 218.

As shown in Fig. 2b, the composer 224 is configured to compose said z syntax elements in stages from the incoming source symbols s1 to sx, x being any 1 to 3 in the present example. There may be two or more stages. As shown in Fig. 2b, the composer 224 preliminarily sets z as the first symbol s1 and checks whether z equals the first limit limit1. If this is not the case, z has been found. Otherwise, composer 224 adds the following source symbol s2 from source symbol stream 218 to z and checks again whether this z equals limit2. If not, z has been found. If not, the composer 224 adds the following source symbol s3 from the source symbol stream 218 to z, to obtain z in its final form. Generalizations about a larger or smaller maximum number of source symbols are easily deducible from the above description, and such alternatives will also be described below.

In any case, all these different components or resulting or source symbols are in accordance with the following embodiments, coded among the coding alternatives. At least one of them is sent by the subdivider to the PIPE encoder 104, and at least one of them is sent to the VLC 102 encoder.

Particular advantageous embodiments are described in more detail below. These details focus on the favorable possibilities of dividing the range of values of the syntax elements and the entropy VLC and PIPE encoding schemes that can be used to encode the source symbols.

Additionally, as also described above with respect to Fig. 1, the entropic decoding apparatus of Fig. 2a can be configured to receive the first bit stream 206 as well as the plurality of second bit streams 216 separately or in an interleaved shape by means of an interleaved bitstream 228. In the latter case, the entropic decoding apparatus of Fig. 2a may comprise a deinterleaver 230 configured to deinterleale the interleaved bitstream 228 to obtain the first bitstream 206 on one side and the plurality of second bitstreams 216 on the one hand other side. Similar to the explanation above in Fig. 1, deinterleaver 230 can be subdivided into two stages, namely deinterleaver 232 for deinterleaving bitstream 228 interleaved into two parts, specifically bitstream 206 on one side and one interleaved form 234 of the second bitstream 216 on the other hand, and a deinterleaver 236 for deinterleaving of the latter bitstream 234 to obtain the individual bitstreams 216.

Thus, Fig. 1a and Fig. 2a showed an entropic encoding apparatus on the one hand and an entropic decoding apparatus suitable for decoding the encoding result obtained by the entropic encoding apparatus of Fig. 1, on the other hand. . Details regarding many of the elements shown in Fig. 1a and the Fig. 2 are described in more detail in relation to additional figures. Accordingly, these details are referred to in the description that follows, and these details should be considered to apply to the Fig. 1a and Fig. 2 embodiments individually as well, provided that these details can be implemented separately in the above embodiments. described. Simply with regard to interleavers and deinterleavers 132 and 234, some further indication is made here. In particular, the interleaving of bit streams 112 and 118 may be favorable if the bit streams are to be multiplexed into a channel to be transmitted. In this case, it may be favorable to interleaving the VLC bitstream 112 on the one hand and the PIPE encoding bitstreams 118 on the other hand so that they obey certain conditions to be satisfied such as obeying some maximum decoding delay. In other words, it may be necessary for the relative time offset between the times of the syntax elements and source symbols, respectively, to be recoverable on the one hand decoding side and the relative time offset according to their position in the bitstream that can be parsed on the other hand, do not exceed a certain maximum delay. Many alternatives to solve this problem are described below. One such possibility involves that entropic encoders 116 be a type of variable length encoder configured to map sequences of alphabet symbols to codewords, and that entropic decoders 210 do the reverse mapping. The code words for the VLC 112 bit stream and the PIPE 118 bit streams can be selected, but do not have to be selected, so that no code word for any of these bit streams is prefixed for any code word for any of the other bit streams, so that the edges of the codewords remain uniquely determinable on the decoder side. In either case, the interleaver 128 may be configured to reserve and buffer a sequence of codeword entries for the codeword within the first bitstream 112 and the second bitstream 118 in sequential order depending on a order in which the alphabet symbols in sequence 124 of alphabet symbols sent by selector 120 to the plurality of entropic encoders 116 result in a start of a new sequence of alphabet symbols to be mapped to a respective code word in the respective entropic encoder 116 and a new source symbol of the second subflow 108 is mapped by the encoder ^v L ^c 102, respectively. In other words, the interleaver 128 inserts the code words from the bit stream 112 into the common bit stream 126 in the order of the source symbols from which they were derived by VLC encoding, in their order within the subflow 108 and source symbol stream 106, respectively. Codewords produced by entropic encoders 116 are inserted into bitstream 126 common among consecutive codewords in bitstream VLC 112. Due to the categorization of the PIPE encoding of alphabet symbols by the allocator 114 and selector 120, respectively, each of the code words of the entropy encoders 116 have alphabet symbols of different source symbols of the subflow 110 encoded therein. The position of the code words of the PIPE encoded bitstreams 118 within the common bitstream 126 and relative to the VLC codeword of the bitstream 112, is determined by the first encoded alphabet symbol in each code word, respectively, ie the oldest in time. The order of these primary alphabet symbols encoded within the codewords of the bitstreams 118 in the alphabet symbol stream 124 determines the order of the codewords of the bitstreams 118 within the bitstream 126 Common to each other, relative to the VLC codewords in bit stream 112, the source symbol to which these primary alphabet symbols coded within the bitstream codewords 118 belong, determine which consecutive codewords in the stream bit 112 the respective code word of any of the bit streams 118 is to be positioned. In particular, the consecutive VLC codewords among which the respective codewords of any of the bitstreams 118 are to be positioned , are those among which the source symbol of subflow 110 is positioned according to the original order of a non-subdivided source symbol stream 106, to which the symbol belongs respective primary alphabet bolus encoded in the respective codeword of bit stream 118. Interleaver 128 can be configured to remove the codewords entered within the aforementioned codeword entries in sequential order to obtain bitstream 126 common interspersed codeword. As already described above, the entropic encoders 116 can be configured to sequentially enter their codewords into the reserved codeword entries for the respective entropic encoder 116, and selector 120 can be configured to output the alphabet symbols representing the symbols. source of the second subflow 110 in an order that maintains an order in which the source symbols of the first subflow 108 and of the second subflow 110 were interleaved within the source symbol sequence 106.

Additional measures can be provided to deal with situations where certain of the entropic encoders 116 are selected so randomly that it takes a long time to obtain a valid codeword within that rarely used entropic encoder 116. Examples of such measures are described in more detail below. In particular, the interleaver 128 together with the entropic encoder 116 can, in this case, be configured to purge their alphabet symbols collected so far and the code words that have been entered into the aforementioned code word entries, respectively , such that the purge procedure time can be predicted or emulated on the decoding side.

On the decoding side, the deinterleaver 230 can act in the reverse direction: provided that, according to the aforementioned analysis scheme, the next source symbol to be decoded is a VLC-encoded symbol, a current code word within the common 228 bit stream is considered as a word of VLC code and is sent within bitstream 206 to VLC decoder 200. On the other hand, as long as any of the alphabet symbols belonging to any of the PIPE encoded symbols in subflow 208 is a primary alphabet symbol, ie need a new mapping of a codeword from a respective one of bit streams 216 to a respective alphabet symbol sequence by the respective entropic decoder 210, the current codeword of common bitstream 228 is considered as a PIPE encoded code and sent to the respective entropic decoder 210. Detecting the next codeword edge, i.e. detecting the extension of the next codeword from the end of the codeword that has just been sent to any of the decoders 200 and 202, respectively, to its end within the incoming interleaved bitstream 228 may be deferred, and performed under the knowledge that decoder 200 and 202 is the specialized receiver of this next codeword according to the rule described above: based on this knowledge, the codebook used by the receiving decoder is known and the respective code word detectable. If, on the other hand, the codebooks were designed so that the edges of the codeword were detectable without a priori knowledge of the receiving decoder between 200 and 202, then codeword separation could be performed in parallel. In any case, due to interleaving, the source symbols are available in the decoder in a decoded entropic form, ie as source symbols, in their correct order with reasonable delay.

Having described the above embodiments for an entropic encoding apparatus and the respective entropic decoding apparatus, the following describes more detailed embodiments of the aforementioned PIPE encoders and PIPE decoders.

A PIPE encoder according to one embodiment is illustrated in Fig. 3. It can be used as a PIPE encoder in Fig. 1a. The PIPE encoder losslessly converts a stream of source symbols 1 to a set of two or more partial bit streams 12. Each source symbol 1 can be associated with a category or type among a set of one or more categories or types. As an example, categories can specify the type of font symbol. In the context of hybrid video encoding, a separate category can be associated with macroblock encoding modes, block encoding modes, reference image indices, motion vector differences, subdivision flags, code block flags, parameters quantification, transform coefficient levels, etc. In other application areas such as audio, voice, text, documents, or data encoding in general, different categorizations of source symbols are possible.

In general, each source symbol can take a value from an infinite or countable set of values, where the set of possible values of the source symbol can differ for different categories of source symbols. To reduce the complexity of the encoding and decoding algorithm and to allow a general encoding and decoding design for different source symbols and source symbol categories, source symbols 1 are converted to ordered sets of binary decisions and these binary decisions are processed below using simple binary coding algorithms. Therefore, the binary converter 2 bijectively maps the value of each source symbol 1 into a sequence (or chain) of binaries 3. Binary sequence 3 represents a set of ordered binary decisions. Each binary 3 or binary decision can take a value from a set of two values, for example, one of the values 0 and 1. The binary conversion scheme may be different for different categories of source symbols. The binary conversion scheme for a particular source symbol category may depend on the set of possible values of the source symbol and / or other properties of the source symbols for the particular category. Table 1 illustrates three examples of binary conversion schemes for infinite accounting sets. The binary conversion schemes for countable infinite sets can also be applied for finite sets of symbol values. In particular for large finite sets of symbol values, inefficiency (resulting from unused binary sequences) may be negligible, but the universality of such binary conversion schemes provides an advantage in terms of complexity and memory requirements. For small finite sets of symbol values, it is often preferable (in terms of coding efficiency) to adapt the binary conversion scheme to the number of possible symbol values. Table 2 illustrates three example binary conversion schemes for finite sets of 8 values. Binary conversion schemes for finite sets can be deduced from universal binary conversion schemes for countable infinite sets by modifying some binary sequences in a way that finite sets of binary sequences represent code without redundancy (and potentially rearranging binary sequences). As an example, the truncated unary binary conversion scheme in Table 2 was created by modifying the binary sequence for source symbol 7 of conversion to universal unary binary (see Table 1). Conversion to truncated and rearranged Exp-Golomb binary of order 0 in Table 2 was created by modifying the sequence of binaries for source symbol 7 of conversion to universal unary binary Exp-Golomb of order 0 (see Table 1) and by reordering the binary sequences (the truncated binary sequence for symbol 7 was assigned to symbol 1). For finite sets of symbols, it is also possible to use unsystematic / non-universal conversion to binary schemes, as exemplified in the last column of Table 2.

Table 1: Examples of conversion to binary for countable infinite sets (or large finite sets).

continuation

Each binary 3 of the binary sequence created by converter to binary 2 is supplied in parameter mapper 4 in sequential order. The parameter mapper assigns a set of one or more parameters to each binary 3 and produces the output of the binary with the associated parameter set 5. The set of parameters is determined in exactly the same way in the encoder and decoder. The parameter set can consist of one or more of the following parameters:

- a measure of an estimate of the probability of one of the two possible binary values for the current binary,

- a measure of a probability estimate for the least likely or most likely binary value for the current binary,

- an identifier that specifies an estimate of which of the two possible binary values represents the most likely or least likely binary value for the current binary,

- the category of the associated source symbol,

- a measure of the importance of the associated source symbol,

- a measure of the location of the associated symbol (eg in temporal, spatial or volumetric data sets),

- an identifier that specifies the protection of the channel code for the binary or associated source symbol, - an identifier that specifies the encryption scheme for the binary or associated source symbol,

- an identifier that specifies a class for the associated symbol,

- the binary number in the binary sequence for the associated source symbol.

In one embodiment, the parameter mapper 4 associates each binary 3, 5 with a measure of a probability estimate for one of the two possible binary values for the current binary. Parameter mapper 4 associates each binary 3, 5 with a measure of a probability estimate for the least likely or most likely binary value for the current binary and an identifier that specifies an estimate of which of the two possible binary values represents the least likely or most likely binary value for the current binary. It should be noted that the probability for the least likely or most probable binary value and the indicator specifying which of the two possible binary values represents the least probable or most probable binary value are equivalent measures of the probability of one of the two possible binary values.

In a further embodiment, the parameter mapper 4 associates each binary 3, 5 with a measure of a probability estimate for one of the two possible binary values for the current binary and one or more additional parameters (which may be one or more than the parameters listed above). In a further embodiment, the parameter mapper 4 associates each binary 3, 5 with a measure of the probability estimate of the least likely or most probable binary value for the current binary, an identifier that specifies the estimate for which of the two possible binary values represents the least likely or most binary value probable for the current binary, and one or more additional parameters (which may be one or more of the parameters listed above).

In one embodiment, the parameter mapper 4 determines one or more of the above-mentioned probability measures (measure of an estimate of the probability of one or more possible binary values for the current binary, measure of an estimate of the probability of the value binary least likely or most likely for current binary, identifier specifying an estimate of which of the two possible binary values represents the least likely or most likely binary value for the current binary) based on a set of one or more symbols already coded. The coded symbols that are used to determine the probability measures may include one or more already coded symbols from the same symbol category, one or more already coded symbols from the same symbol category that correspond to the data set (such as blocks or groups of samples) from neighboring spatial and / or temporal locations (relative to the data set associated with the current source symbol), or one or more symbols already coded from different symbol categories that correspond to data sets of the same and / or or neighboring spatial and / or temporal locations (relative to the dataset associated with the current source symbol).

Each binary with an associated set of parameters 5 that is output from parameter mapper 4 is fed to a binary buffer selector 6. Binary buffer selector 6 potentially modifies the value of input binary 5 based on the value of the input binary and associated parameters 5 and feeds the output binary 7 - with a potentially modified value - to one of two or more binary 8 buffers. The binary 8 buffer to which the output binary is sent 7 is determined based on the value of input binary 5 and / or the value of associated parameters 5.

In one embodiment, the binary buffer selector 6 does not modify the binary value, that is, the output binary 7 always has the same value as the input binary 5.

In a further embodiment, the binary buffer selector 6 determines the value of the output binary 7 based on the input binary value 5 and the associated measure for a probability estimate for one of the two possible binary values for the current binary. In one embodiment, the value of the output binary 7 is set equal to the value of the input binary 5 if the probability measure for one of the two possible binary values for the current binary is less than (or less than or equal to ) a particular threshold; if the probability measure for one of the two possible binary values for the current binary is greater than or equal to (or greater than) a particular threshold, the output binary 7 value is modified (i.e. set opposite to the value of the input binary). In a further embodiment, the value of the output binary 7 is set equal to the value of the input binary 5 if the probability measure for one of the two possible binary values for the current binary is greater than (or greater than or equal to) a) a particular threshold; if the measure of the probability of one of the two possible binary values for the current binary is less than or equal to (or less than) a particular threshold, the output binary value 7 is modified (i.e. set opposite to the value of the input binary). In one embodiment, the threshold value corresponds to a value of 0.5 for the estimated probability of both possible binary values.

The binary buffer selector 6 can determine the value of the output binary 7 based on the value of the input binary 5 and the associated identifier that specifies an estimate for which of the two possible binary values represents the least likely binary value. or more likely for the current binary. In one embodiment, the output binary value 7 is set equal to the input binary value 5 if the identifier specifies that the first of the two possible binary values represents the least likely (or most likely) binary value for the binary current, and the output binary value 7 is modified (i.e. set to the opposite of the input binary value) if the identifier specifies that the second of the two possible binary values represents the least likely binary value (or more probable) for the current binary.

The binary buffer selector 6 can determine the binary buffer 8 to which the output binary 7 is sent based on the associated measure for an estimate of the probability for one of the two possible binary values for the current binary . In one embodiment, the set of possible values for the measurement of a probability estimate for one of the two possible binary values is finite and the binary buffer selector 6 contains a table that exactly associates a binary buffer 8 with each possible value for estimating the probability of one of the two possible binary values, where different values for the measurement of an estimation of the probability of one of the two possible binary values can be associated with the same binary buffer 8. In a further embodiment, the range of possible values for the measurement of a probability estimate for one of the two possible binary values is partitioned into a number of intervals, the binary buffer selector 6 determines the index of the interval for the current measurement for an estimate of the probability for one of the two possible binary values, and the memo selector Binary buffer 6 contains a table that exactly associates a binary 8 buffer with every possible value for the interval index, where different values for the interval index can be associated with the same binary 8 buffer. In a preferred embodiment , input binaries 5 with opposite measures for an estimate of the probability for one of the two possible binary values (measure opposite to that representing probability estimates P and 1 - P), are fed to the same binary buffer 8. In a further preferred embodiment, associating the measure for an estimate of the probability for one of the two possible binary values for the current binary with a particular binary buffer is adapted over time, for example to ensure that the created partial bit streams have similar bit rates.

The binary buffer selector 6 can determine the binary buffer 8 to which the output binary 7 is sent based on the associated measure for an estimate of the probability for the least likely or most probable binary value for the binary current. In a preferred embodiment, the set of possible values for the measurement of a probability estimate for the least likely or most likely binary value is finite and the binary buffer selector 6 contains a table that exactly associates a buffer of binary 8 with each possible value of the probability estimate for the least likely or most likely binary value, where different values for the measure of an estimate of probability for the least likely or most likely binary value can be associated with it Binary buffer 8. In a further embodiment, the range of possible values for the measurement of a probability estimate for the least likely or most likely binary value is partitioned over a number of intervals, the binary buffer selector 6 determines the index of the interval for the current measure of an estimate of the probability for the binary value minus proba ble or more likely, and the binary buffer 6 selector contains a table that exactly associates a binary buffer 8 with each possible value for the interval index, where different values for the interval index can be associated with the same memory binary buffer 8. In a further embodiment, associating the measure of a probability estimate for the least likely or most likely binary value for the current binary with a particular binary buffer is tailored over time, for example to ensure that created partial bit streams have similar bit rates.

Each of the two or more binary buffers 8 connects to exactly one binary encoder 10 and each binary encoder connects only to one binary buffer 8. Each binary encoder 10 reads binaries from the binary buffer 8 associates and converts a sequence of binaries 9 into a code word 11, which represents a sequence of bits. Binary 8 buffers represent first-in-first-out buffers; binaries that are fed later (in sequential order) into binary buffer 8 are not encoded before binaries that are fed before (in sequential order) into binary buffer. Codewords 11 that are output from a particular binary encoder 10 are written to a particular partial bitstream 12. The global encoding algorithm converts the source symbols 1 into two or more partial bitstreams 12, where the number of partial bitstreams is equal to the number of binary buffers and binary encoders. In a preferred embodiment a binary encoder 10 converts a variable number of binaries 9 into a code word 11 of a variable number of bits. An advantage of the PIPE encoding embodiments described above and below is that the binary encoding can be performed in parallel (for example, for different sets of probability measures), which reduces processing time for various implementations.

Another advantage of PIPE encoding is that binary encoding, which is performed by binary encoders 10, can be specifically designed for different sets of parameters 5. In particular, binary encoding and decoding can be optimized (in terms of encoding efficiency and / or complexity) for different groups of estimated probabilities. On the other hand, this allows a reduction of the coding / decoding complexity in relation to the arithmetic coding algorithms with similar coding efficiency. On the other hand, it allows an improvement of the coding efficiency in relation to the VLC coding algorithms with similar coding / decoding complexity. In one embodiment, the binary encoders 10 implement different encoding algorithms (i.e. mapping binary sequences to codewords) for different sets of measurements for an estimate of the probability for one of the two possible binary values 5 for the current binary. In a further embodiment, the binary encoders 10 implement different encoding algorithms for different sets of measurements for an estimate of the probability for the least likely or most probable binary value for the current binary. In a further embodiment, the binary encoders 10 implement different encoding algorithms for different channel protection codes. In a further embodiment, the binary encoders 10 implement different encoding algorithms for different encryption schemes. In a further embodiment, the binary encoders 10 implement different encoding algorithms for different combinations of channel protection codes and measurement groups for an estimate of the probability for one of the two possible binary values 5 for the current binary. In a further embodiment, the binary encoders 10 implement different encoding algorithms for different combinations of channel protection codes and groups of measurements for an estimate of the probability for the binary value 5 less likely or more probable for the current binary. In a further embodiment, the binary encoders 10 implement different encoding algorithms for different combinations of encryption schemes and groups of measures for an estimate of the probability for one of the two possible binary values 5 for the current binary. In a further embodiment, the Binary encoders 10 implement different encoding algorithms for different combinations of encryption schemes and groups of measures for an estimate of the probability for the least probable or most probable binary value 5 for the current binary.

The binary encoders 10 - or one of the binary encoders - can represent binary arithmetic encoding engines. In a further embodiment, one or more of the binary encoders represent a binary arithmetic encoding engine, in which the mapping from the representative LPS / LPB probability plps of a given binary buffer to a corresponding code interval width Rlps - that is, the interval subdivision of the internal state of the binary arithmetic coding engine, which is defined by the width of the current interval R and the offset of the current interval L, which identifies, for example, the lower bound of the code interval - It is done through the use of a lookup table. In a further preferred embodiment, for each table-based binary arithmetic coding engine associated with a given binary buffer, K values representative of the interval width {Qo, ..., Q ^k - ⁱ } are used for the representation of R ^lps with the choice of K and the representative values of the interval width {Qo, ..., Q ^k - ⁱ } dependent on the binary buffer. For a choice of K > 1, arithmetic coding of a binary may involve mapping sub-steps of the current interval width R to a quantization index q with values of {0, ..., K-1} and perform the subdivision of intervals by accessing the value of the corresponding partial interval width Qq from a lookup table using q as an index. For a choice of K = 1, that is, for the case where only a representative interval width value Qo is given, this Qo value may be chosen as a power of two to allow decoding of multiple MPS / MPB values that they enter the corresponding binary buffer within a single normalization cycle. The resulting codewords for each arithmetic coding engine can be transmitted separately, in packets, or stored, or can be interleaved for the purpose of transmission or storage as described herein below.

That is, a binary arithmetic encoding engine 10 could perform the following steps in encoding the binaries into its binary 8 buffer:

1. Receive valLPS, binaries from the binary buffer (reminder: the respective 10 binary arithmetic encoding engine considered here has been chosen to receive "binary" (or, in other words, "binary" is associated with the encoding engine respective binary arithmetic 10) because the estimate of the probability distribution, such as p_state [binary], is associated with that binary arithmetic coding engine 10)

2. Quantification of R:

q_index = Qtab [R >> q] (or some other form of quantification)

(or some other form of quantification)

3. Determination of RLPsy R:

Rlps = Rtab [q_index] (note that p_status has not been mentioned here, since it is set for the binary arithmetic coding engine 10 considered, ie p_status [encoder], and Rtab has stored in it the calculated values for p [ p_status [encoder]] Q [q_index] R = R - Rlps [that is, R is preliminarily pre-updated as if "binary" were MPS]

4. Calculation of the new partial interval:

if (binary = I - valMPS) then

L ^ L R

R ^ Rlps

5. Renormalization of L and R, writing bits,

in which

q_index describes the index of a quantized value read from Qtab,

p_status describes the current state (set for binary arithmetic coding engine 10),

Rlps describes the interval width corresponding to LPS and

valMPS describes the bit value that corresponds to the MPS.

Accordingly, a binary arithmetic decoding engine 22 could perform the following steps in decoding the binary buffer 20 output binaries:

1. Receive the request for a binary (reminder: the respective binary 22 arithmetic decoding engine considered here has been chosen to decode “binary” (or, in other words, “binary” is associated with the respective binary arithmetic decoding engine 22) because the estimate of the probability distribution, such as p_state [binary], is associated with that binary arithmetic decoding engine 22)

2. Quantification of R:

q_index = Qtab [R >> q] (or some other form of quantification)

(or some other form of quantification)

3. Determination of RlpsY R:

Rlps = Rtab [q_index] (note that p_status has not been mentioned here, since it is set for the binary arithmetic decoding engine 22 considered, that is p_status [encoder], and Rtab has stored in it the pre-calculated values for p [p_state [encoder]] Q [q_index]

R = R - Rlps [that is, R is preliminary pre-updated as if "binary" were MPS]

4. Determination of binary depending on the position of the partial interval:

if (V> R) then

binary ^ 1 - valMPS (binary is decoded as LPS; binary buffer selector 18 will get the actual binary value by using this binary and valMPS information)

V ^ V -R

R ^ Rlps

otherwise

binary ^ valMPS (binary is decoded as MPS; binary buffer selector 18 will get the actual binary value by using this binary and valMPS information)

5. Renormalization of R, reading a bit and updating V,

in which

q_index describes the index of a quantized value read from Qtab,

p_status describes the current state (set for binary arithmetic decoding engine 22),

Rlps describes the interval width corresponding to the LPS,

valMPS describes the value of the bit that corresponds to the MPS, and V describes a value from within the current partial interval.

Binary encoders 10 - or one or more of the binary encoders - can represent entropy encoders that directly map sequences of input binaries 9 over codewords 10. Such mappings can be implemented efficiently and do not require a complex arithmetic encoding engine. Reverse mapping of code words over binary sequences (as performed in the decoder) should be unique to ensure perfect decoding of the input sequence, but mapping of binary 9 sequences over code words 10 does not necessarily require being unique, that is, it is possible that a particular sequence of binaries can be mapped over more than one sequence of codewords. In one embodiment, the mapping of input binary sequences 9 to codewords 10 is bijective. In a further preferred embodiment, the binary encoders 10 - or one or more of the binary encoders - represent entropy encoders that directly map variable length sequences of input binaries 9 into variable length codewords 10. In one embodiment, exit code words represent non-redundant codes such as general Huffman codes or canonical Huffman codes.

Two examples of bijective mapping of binary sequences to codes without redundancy are illustrated in Table 3. In a further embodiment, the exit code words represent redundant codes suitable for error detection and error recovery. In a further preferred embodiment, the exit code words represent encryption codes suitable for encryption of the source symbols.

T l: E m l m nr n i in ri l r i.

In a further embodiment, the binary encoders 10 - or one or more of the binary encoders - represent entropy encoders that directly map variable length sequences of input binaries 9 into fixed length codewords 10. In a further embodiment, binary encoders 10 - or one or more of the binary encoders - represent entropy encoders that directly map fixed-length sequences of input binaries 9 into variable-length codewords 10.

A PIPE decoder according to an embodiment is illustrated in Figure 4. The decoder basically performs the inverse operations of the encoder of Fig. 3, so that the source symbol sequence 27 (previously encoded) is decoded from a set of two or more partial 24 bit streams. The decoder includes two different process streams: a stream for data requests, which replicates the encoder's data stream, and a data stream that represents the inverse of the encoder's data stream. In the illustration in Fig. 4, the dashed arrows represent the data request flow, while the solid arrows represent the data flow. The decoder building blocks basically replicate the encoder building blocks, but implement the inverse operations.

Decoding a source symbol is triggered by a request for a new decoded source symbol 13 that is sent to the binary converter 14. In one embodiment, each request for a new decoded source symbol 13 is associated with a category of a set of a or more categories. The category that is associated with a request for a source symbol is the same as the category that was associated with the corresponding source symbol during encoding.

Converter to binary 14 maps the request for a source symbol 13 into one or more requests for a binary that is sent to parameter mapper 16. As the final response to a request for a binary that is sent to parameter mapper 16 by the converter to binary 14, converter to binary 14 receives a decoded binary 26 from binary buffer selector 18. Converter to binary 14 compares the received sequence of decoded binaries 26 with the binary sequences of a conversion scheme to particular binary for the requested source symbol and, if the received sequence of decoded binaries 26 matches the conversion to binary of a source symbol, the binary converter empties its binary buffer and produces the output of decoded source symbols as a final response to the request for a new decoded symbol. If the already received sequence of decoded binaries does not match any of the binary sequences in the binary conversion scheme for the requested source symbol, the binary converter sends another request for a binary to the parameter mapper until the decoded binary sequence match one of the binary sequences in the binary conversion scheme for the requested source symbol. For each request for a source symbol, the decoder uses the same binary conversion scheme that was used for encoding the corresponding source symbol. The binary conversion scheme may be different for different categories of source symbols. The binary conversion scheme for a particular source symbol category may depend on the set of possible source symbol values and / or other properties of the source symbols for the particular category.

The parameter mapper assigns a set of one or more parameters to each request for a binary and sends the request for a binary with the associated set of parameters to the binary buffer selector. The set of parameters that are assigned to the binary requested by the parameter mapper is the same as that set assigned the corresponding binary during encoding. The parameter set may consist of one or more of the parameters that were mentioned in the encoder description.

Parameter mapper 16 can associate each request for a binary with a measure of an estimate of the probability of one or more of the possible binary values for the current requested binary. In particular, parameter allocator 16 can associate each request for a binary with a measure of an estimate of the probability of the least likely or most probable binary value for the current requested binary and an identifier that specifies an estimate for which of the two Possible Binary Values represents the least likely or most likely binary value for the current requested binary.

In a further embodiment, the parameter mapper 16 associates each request for a binary 15, 17 with a measure of a probability estimate for one of the two possible binary values for the current requested binary and one or more additional parameters. In a further preferred embodiment, the parameter mapper 16 associates each request for a binary 15, 17 with a measure of a probability estimate for the least likely or most likely binary value for the current requested binary, an identifier that specifies a estimate for which of the two possible binary values represents the least likely or most probable binary value for the current requested binary and one or more additional parameters (which may be one or more of the parameters listed above).

In one embodiment, the parameter mapper 16 determines one or more of the aforementioned probability measures (measure of a probability estimate for one of the two possible binary values for the current requested binary, measure for a probability estimate for the least likely or most likely binary value for the current requested binary, identifier that specifies an estimate for which of the two possible binary values represents the least likely or most likely binary value for the current requested binary) based on a set of one or more already decoded symbols. Determining the probability measures for a particular request for a binary replicates the process in the encoder for the corresponding binary. Decoded symbols that are used for the determination of probability measures may include one or more already decoded symbols of the same symbol category, one or more already decoded symbols of the same symbol category that correspond to data sets (such as blocks or groups of samples) from neighboring spatial and / or temporal locations (relative to the data set associated with the current request for a source symbol), or one or more already decoded symbols from different symbol categories corresponding to data sets from the same and / or neighboring spatial and / or temporal locations (relative to the dataset associated with the current request for a source symbol).

Each request for a binary with an associated set of parameters 17 which is the output of parameter mapper 16 is provided to a binary buffer selector 18. Based on the associated parameter set 17, the binary buffer selector 18 sends a request for a binary 19 to one of two or more binary buffers 20 and receives a decoded binary 25 from the selected binary buffer 20. The decoded input binary 25 is potentially modified and the decoded output binary 26 - potentially with a modified value - is sent to the binary converter 14 as a final response to the request for a binary with an associated set of parameters 17.

The binary buffer 20 to which the request for a binary is sent is selected in the same way as the binary buffer to which the output binary of the binary buffer selector on the encoder side was sent.

In one embodiment, the binary buffer selector 18 determines the binary buffer 20 to which the request for a binary 19 is sent based on the associated measure for an estimate of the probability for one of the two possible binary values. for the current requested binary. In one embodiment, the set of possible values for the measurement of a probability estimate for one of the two possible binary values is finite and the binary buffer selector 18 contains a table that exactly associates a binary buffer 20 with each possible value of the probability estimate for one of the two possible binary values, in which different measurement values of a probability estimate for one of the two possible binary values can be associated with the same buffer binary 20. In one embodiment, the range of possible values for the measurement of a probability estimate for one of the two possible binary values is partitioned over a number of intervals, the binary buffer selector 18 determines the index of the interval for the current measurement of an estimate of the probability for one of the two possible binary values, and the internal memory selector Binary mean 18 contains a table that exactly associates a binary buffer 20 with each possible value for the interval index, where different values for the interval index can be associated with the same binary buffer 20. Requests for binaries 17 with opposite measures of a probability estimate for one of the two possible binary values (opposite measures are those representing probability estimates P and 1 - P) can be sent to the same binary buffer 20. Additionally, it can be adapted to over time the association of the measure of an estimate of the probability for one of the two possible values of binary for the current binary request with a particular binary buffer.

The binary buffer selector 18 can determine the binary buffer 20 to which the request for a binary 19 is sent based on the associated measure of a probability estimate for the least likely or most likely binary value for the current requested binary. In one embodiment, the set of possible values for the measurement of a probability estimate for the least likely or most likely binary value can be finite and the binary buffer selector 18 contains a table that exactly associates a buffer of binary 20 with each possible probability estimate value for the least likely or most likely binary value, where different values for the measurement of a probability estimate for the least likely or most likely binary value may be associated with the same Binary buffer 20. In one embodiment, the range of possible values for the measurement of a probability estimate for the least likely or most likely binary value is partitioned over a number of intervals, the binary buffer selector 18 determines the interval index for the current measure of a probability estimate for the least likely or most binary value probable, and the binary buffer selector 18 contains a table that exactly associates a binary buffer 20 with each possible value for the interval index, where different values for the interval index can be associated with the same binary buffer 20. In a further preferred embodiment, associating the measurement of a probability estimate for the least likely or most likely binary value for the current binary request with a particular binary buffer is tailored over time.

After receiving a decoded binary 25 from the selected binary buffer 20, the binary buffer selector 18 potentially modifies the input binary 25 and sends the output binary 26 - with a potentially modified value - to the binary converter 14 The binary input / output mapping of the binary buffer selector 18 is the inverse of the binary input / output mapping of the binary buffer selector on the encoder side.

The binary buffer selector 18 can be configured not to modify the binary value, that is, the output binary 26 always has the same value as the input binary 25.

The binary buffer selector 18 can determine the value of the output binary 26 based on the value of the input binary 25 and the measurement of a probability estimate for one of the two possible binary values for the current requested binary that is associated with the request for a binary 17. In one embodiment, the value of the output binary 26 is set equal to the value of the input binary 25 if the measure of the probability for one of the two possible binary values for the request for current binary is less than (or less than or equal to) a particular threshold; if the probability measure for one of the two possible binary values for the current binary request is greater than or equal to (or greater than) a particular threshold, the value of the output binary 26 is modified (i.e., it is fixed opposite the input binary value). In a further embodiment, the value of the output binary 26 is set equal to the value of the input binary 25 if the probability measure for one of the two possible binary values for the current binary request is greater than (or greater than or equal to) a particular threshold; if the probability measure for one of the two possible binary values for the current binary request is less than or equal to (or less than) a particular threshold, the value of the output binary 26 is modified (i.e., it is sets to the opposite of the input binary value). In a preferred embodiment, the threshold value corresponds to a value of 0.5 for the estimated probability for both possible binary values.

In a further embodiment, the binary buffer selector 18 determines the value of the output binary 26 based on the value of the input binary 25 and the identifier, which specifies the estimation of which of the two possible binary values the value represents. less likely or more likely binary for the current binary request, which is associated with the request for a binary 17. In a preferred embodiment, the value of the output binary 26 is set equal to the input binary value 25 if the identifier specifies that the first of two possible binary values represents the least likely (or most likely) binary value for the current binary request, and the output binary value 26 is modified (i.e. set to the opposite of the value of the input binary) if the identifier specifies that the second of the two possible binary values represents the least likely (or most likely) binary value for the current binary request l.

As described above, the binary buffer selector sends a request for a binary 19 to one of the two or more binary buffers 20. Binary buffers 20 represent first-in-first-out buffers, which are they feed decoded binary streams 21 from the connected binary decoders 22. In response to a request for a binary 19 that is sent to a binary buffer 20 from the binary buffer selector 18, the binary buffer 20 removes from its content the binary that was first fed into the binary 20 and sends it to binary buffer selector 18. Binaries that were previously sent to binary buffer 20 are removed earlier and sent to binary buffer selector 18.

Each of the two or more binary buffers 20 connects to exactly one decoder. Binary 22 and each binary decoder connects only to a binary buffer 20. Each binary decoder 22 reads code words 23, representing bit streams, from a separate partial bit stream 24. The binary decoder converts a code word 23 into a sequence of binaries 21 that is sent to the connected binary buffer 20. The global decoding algorithm converts two or more partial bit streams 24 into a number of decoded source symbols, where the number of partial bitstreams equals the number of binary buffers and binary decoders and the decoding of source symbols is triggered by requests for new source symbols. In one embodiment, a binary decoder 22 converts code words 23 of a variable number of bits into a sequence of a variable number of binaries 21. An advantage of the above PIPE embodiments is that decoding of binaries from two or more streams of Partial bits can be done in parallel (for example, for different sets of probability measures), reducing processing time for multiple implementations.

Another advantage of the above PIPE decoding embodiments is that the binary decoding, which is performed by the binary decoders 22, can be specifically designed for different sets of parameters 17. In particular, the binary encoding and decoding can be optimized (in terms of coding efficiency and / or complexity) for different groups of estimated probabilities. On the one hand, this allows a reduction of the coding / decoding complexity in relation to arithmetic coding algorithms with similar coding efficiency. On the other hand, it allows an improvement of the coding efficiency in relation to the VLC coding algorithms with similar coding / decoding complexity. In one embodiment, the binary decoders 22 implement different decoding algorithms (i.e. mapping binary streams into codewords) for different sets of measures of a probability estimate for one of the two possible binary values 17 for the current binary request. In a further embodiment, the binary decoders 22 implement different decoding algorithms for different sets of measurements of an estimate of the probability for the least probable or most probable binary value for the current requested binary. In a further embodiment, the binary decoders 22 implement different decoding algorithms for different channel protection codes. In a further embodiment, the binary decoders 22 implement different decoding algorithms for different encryption schemes. In a further embodiment, the binary decoders 22 implement different decoding algorithms for different combinations of channel protection codes and groups of measurements of a probability estimate for one of the two possible binary values 17 for the current requested binary. In a further embodiment, the binary decoders 22 implement different decoding algorithms for different combinations of channel protection codes and groups of measures of an estimate of the probability for the binary value 17 less likely or more likely for the current requested binary . In a further embodiment, the binary decoders 22 implement different decoding algorithms for different combinations of encryption schemes and groups of measures of a probability estimate for one of the two possible binary values 17 for the current requested binary. In a further embodiment, the binary decoders 22 implement different decoding algorithms for different combinations of encryption schemes and groups of measures of an estimate of the probability for the binary value 17 less likely or more probable for the current requested binary.

The binary decoders 22 reverse map the corresponding binary encoders on the encoder side.

In a further embodiment, the binary decoders 22 - or one or more of the binary decoders - represent binary arithmetic decoding engines.

In a further embodiment, the binary decoders 22 - or one or more of the binary decoders - can represent entropy decoders that directly map codewords 23 into binary sequences 21. Such mappings can be implemented efficiently and do not require an encoding engine. complex arithmetic. The mapping of codewords into binary sequences must be unique. In one embodiment, the mapping of code words 23 into binary sequences 21 is bijective. In a further embodiment, the binary decoders 10 - or one or more of the binary decoders - represent entropy decoders that directly map codewords 23 of variable length into binary sequences 21 of variable length. In one embodiment, the input codewords represent non-redundant codes such as general Huffman codes or canonical Huffman codes. Two examples of bijective code mapping without redundancy in binary sequences are illustrated in Table 3. In a further embodiment, the input code words represent redundant codes suitable for error detection and error recovery. In a further embodiment, the input code words represent encrypted codes.

The binary decoders 22 - or one or more of the binary decoders - can represent entropy decoders that directly map fixed-length codewords 23 into variable length binary sequences 21. Alternatively, the binary decoders 22 - or one or more of the binary decoders - represent entropic decoders that directly map variable length codewords 23 into fixed length binary sequences 21.

Thus, Fig. 3 and Fig. 4 show an embodiment for a PIPE encoder for encoding a source symbol sequence 1 and a PIPE decoder for reconstruction thereof. That is, the PIPE encoder of Fig. 3 can be used as the PIPE encoder 104 in Fig. 1a with a binary converter 2 acting as the symbolizer 122, the parameter allocator 4 acting as the allocator 114, the memory selector acting binary buffer 6 as selector 120, and the pair of binary buffers 8 and binary encoder 10 acting in series as one of the respective entropic encoders 116 each producing bitstreams 12 corresponding to the streams of bits 118 in Fig. 1a. As is clear from the comparison of FIG. 3 and FIG. 1, the mapper 114 in FIG. 1a may have its input alternatively connected to the input side of the symbolizer 122 instead of the output side of the latter. Similarly, the PIPE decoder of Fig. 4 can be used as the PIPE decoder 202 of Fig. 2a with partial bitstreams 24 corresponding to bitstreams 216 in Fig. 2, buffer pairs 20 corresponding. and binary decoder 22 connected in series to the individual entropic decoders 210, the binary buffer selector 18 acting as the selector 214, the parameter mapper 16 acting as the mapper 212 and the converter to binary 14 acting as the de-symbolizer 222 Again, a comparison between Fig. 2a and Fig. 4 makes it clear that the interconnection between the de-symbolizer 222, allocator 212 and selector 214 can be configured differently, so that, according to an alternative embodiment, the connections of Fig. 2a are modified to correspond to those shown in Fig. 4.

The PIPE encoder of Fig. 3 comprises an allocator 4 configured to assign a number of parameters 5 to each alphabet symbol in the alphabet symbol sequence 3. The assignment is based on information contained within the previous alphabet symbols in the sequence of alphabet symbols so that the category of syntax element 1 to whose representation - such as conversion to binary - the current alphabet symbol belongs and which, according to the syntax structure of syntax elements 1, is expected currently, an expectation which, in turn, is deducible from the history of previous syntax elements 1 and alphabet symbols 3. Additionally, the encoder comprises a plurality of entropic encoders 10 each of which is configured to convert alphabet symbols 3 sent to the respective entropic encoder in a respective bitstream 12, and a selector 6 configured to send each alphabet symbol 3 to one selected from the plurality of entropic encoders 10, the selection of the number of parameters 5 assigned to the respective alphabet symbol 3 depending. The PIPE decoder of Fig. 4 comprises a plurality of entropic decoders 22, each of which is configured to convert a respective bit stream 23 in alphabet symbols 21; an allocator 16 configured to assign a number of parameters 17 to each alphabet symbol 15 of a sequence of alphabet symbols to be reconstructed based on information contained within previously reconstructed alphabet symbols of the sequence of alphabet symbols (see 26 and 27 in Fig. 4); and a selector 18 configured to retrieve each alphabet symbol from the sequence of alphabet symbols to be reconstructed from one selected from the plurality of entropic decoders 22, the selection depending on the number of parameters defined for the respective alphabet symbol. The allocator 16 can be configured such that the number of parameters assigned to each alphabet symbol comprises, or is, a measure of an estimate of a probability of distribution among the possible alphabet symbol values that a respective alphabet symbol can assume. The sequence of alphabet symbols to be reconstructed can be from a binary alphabet and allocator 16 can be configured such that the estimate of the probability distribution consists of a measure of an estimate of a probability of a less likely or more likely binary value of the two possible binary values of the binary alphabet and an identifier that specifies the estimation of which of the two possible binary values represents the least probable or most probable binary value. Mapper 16 may be further configured to internally assign a context to each alphabet symbol in the alphabet symbol sequence 15 to be reconstructed based on the information contained within the previously reconstructed alphabet symbols in the alphabet symbol sequence to be reconstructed having each context a respective probability distribution estimate associated therewith, and adapting the probability distribution estimate for each context to a current symbol statistic based on previously reconstructed alphabet symbol symbol values to which the respective context is assigned. The context can take into account a spatial relationship or neighborhood of positions to which the syntax elements belong such as in video or image encoding, or even in tables in the case of financial applications. Next, the measure of the probability distribution estimate for each alphabet symbol can be determined based on the probability distribution estimate associated with the context assigned to the respective alphabet symbol such as by quantifying the associated probability distribution estimate with the context assigned with the respective alphabet symbol to one of a plurality of representative probability distribution estimates to obtain the measure for the probability distribution estimate. The selector may be configured such that a surjective association is defined between the plurality of entropy encoders and the plurality of representative probability distribution estimates, i.e. each entropic encoder has at least one representative probability distribution estimate associated therewith, but More than one representative probability distribution estimate can be associated with an entropic encoder. According to one embodiment, the association can even be bijective. Selector 18 can be configured to change a quantization mapping from a range of probability distribution estimates to the plurality of probability distribution estimates representative in a predetermined deterministic way depending on previously reconstructed alphabet symbols from the sequence of alphabet symbols, over time. That is, selector 18 can change the sizes of the quantization step, that is, the ranges of probability distributions mapped to individual probability indices which, in turn, can be projectively associated with the individual entropic decoders. The plurality of entropic decoders 22, in turn, can be configured to accommodate their way of converting alphabet symbols into bit streams in response to a change in quantization mapping. For example, each entropy decoder 22 may be optimized for, i.e. may have an optimal compression rate for, a certain probability distribution estimate within the respective probability distribution estimate quantization range, and may change its word mapping of code / symbol sequences so as to adapt the position of this certain probability distribution estimate within the respective probability distribution estimate quantization range after a change of the latter, so as to optimize it. The selector can be configured to change the quantization mapping so that the rates at which the alphabet symbols are retrieved from the plurality of entropic decoders become less scattered. As with the binary converter 14, it is observed that it can be suppressed if the syntax elements are already binary. Additionally, depending on the type of decoder 22, the existence of buffers 20 is not necessary. Additionally, buffers can be integrated into decoders.

So far, more details have been described for the PIPE 104 encoder and the PIPE 202 decoder in Figs. 1a and 2 above, with respect to Figs. 3 and 4, which, if clearly implemented in the apparatus of Figs. 1a and 2, leads to a production of a parallel bitstream in which the VLC and PIPE partial bitstreams are transported in parallel. Possibilities on how to combine the PIPE partial bitstreams to be transmitted are then described below, together with VLC bitstreams in parallel, or secondly interleaving both bitstreams, i.e. the VLC bitstream and the interleaved PIPE bitstream.

Finishing sequences of finite source symbols

In one embodiment for PIPE encoders and decoders, encoding and decoding is performed for a finite set of source symbols. Often a certain amount of data is encoded such as a still image, a frame or field of a video sequence, a fragment of an image, a fragment of a frame or a field of a video sequence, a set of successive audio samples , etc. For finite sets of source symbols, in general, the partial bit streams that are created on the encoder side have to be terminated, i.e. it has to be ensured that all source symbols can be decoded from the transmitted or partial bit streams. stored. After the last binary is inserted into the corresponding binary buffer 8, the binary encoder 10 has to ensure that a complete codeword is written to the partial bitstream 12. If the binary encoder 10 represents a binary arithmetic encoding engine, the arithmetic codeword is to be finalized. Binary encoder 10 represents an entropic encoder that implements direct mapping of binary sequences over codewords, the binary sequence that is stored in the binary buffer after writing the last binary to the binary buffer might not represent a binary sequence that is associated with a codeword (that is, it could represent a prefix of two or more binary sequences that are associated with codewords). In that case, any of the code words associated with a binary sequence containing the binary sequence in the binary buffer as a prefix has to be written to the partial bitstream (the binary buffer has to be purged). This could be done by inserting binaries with a particular or arbitrary value into the binary buffer until a code word is written. In a preferred embodiment, the binary encoder selects one of the codewords with minimum length (in addition to the property that the associated binary sequence must contain the binary sequence in the binary buffer as a prefix). On the decoder side, binary decoder 22 can decode more binaries than required for the last codeword in a partial bitstream; these binaries are not requested by the binary buffer selector 18 and are discarded and ignored. Decoding of the finite set of symbols is controlled by requests for decoded source symbols; if no additional source symbols are requested for a quantity of data, the decoding is finished.

Transmission and multiplexing of partial bit streams

The partial bit streams 12 that are created by the PIPE encoder can be transmitted separately, or can be multiplexed into a single bit stream, or the code words of the partial bit streams can be interleaved into a single bit stream.

In one embodiment, each partial bitstream for a quantity of data is written to a data packet. The amount of data may be an arbitrary set of source symbols such as a still image, a field or frame of a video sequence, a fragment of a still image, a fragment of a field or frame of a video sequence, or a plot of audio samples, etc.

In another preferred embodiment, two or more of the partial bit streams for the data amount or all of the partial bit streams for a data amount are multiplexed into a data packet. The structure of a data packet containing multiplexed partial bit streams is illustrated in Figure 5. That is, the data packet shown in Fig. 5 could be part of intermediate interleaved stream 132 and 234, respectively.

Data packet 300 consists of a header and a partition of the data for each partial bitstream (for the amount of data considered). The data packet header 300 contains indications for partitioning the (remainder of) the data packet into bitstream data segments 302. Along with the partition indications, the header may contain additional information. In one embodiment, the flags for partitioning the data packet are locations of the beginning of the data segments in units of bits or bytes or multiples of bits or multiples of bytes. In one embodiment, the locations of the start of the data segments are encoded as absolute values in the header of the data packet, either relative to the start of the data packet or relative to the end of the header or relative to the start of the data packet. previous data package. In a further embodiment, the locations of the start of the data segments are differentially encoded, that is, only the difference between the actual start of the packet of a data segment and a prediction for the start of the data segment is encoded. The prediction can be deduced based on already known or transmitted information such as the overall size of the data packet, the size of the header, the number of data segments in the data packet, the location of the start of the preceding data segments. In one embodiment, the location of the start of the first data packet is not encoded, but is derived based on the size of the header of the data packet. On the decoder side, the transmitted partition indications are used to deduce the start of the data segments. The data segments are then used as partial bit streams and the data contained in the data segments is provided to the corresponding binary decoders in sequential order.

There are several alternatives for multiplexing the partial bit streams 12 into a data packet. An alternative, which can reduce the requirement for secondary information, particularly for cases where the sizes of the partial bit streams are very similar, is illustrated in Fig. 6. The payload of the data packet, ie data packet 310 without its header 311 is partitioned into segments 312 in a predefined way. As an example, the payload of the data packet can be partitioned into segments of the same size. Each segment is then associated with a partial bitstream or the first part of a partial bitstream 313. If a partial bitstream is greater than the associated data segment, its remainder 314 is placed within the unused space in the end of the other data elements. This can be done in a way that the remaining part of a bit stream is inserted in reverse order (starting from the end of the data segment), which reduces secondary information. The association of the remnants of the partial bitstreams to data segments and, when more than one remainder is added to the data segment, the starting point for one more of the residues has to be signaled within the bitstream, for example in the header of the data packet.

Variable-length codeword interleaving

For some applications, the above-described multiplexing of partial bitstreams 12 (for a number of source symbols) in a data packet may have the following disadvantages: on the one hand, for small data packets, the number of bits of the information Secondary that is required for partitioning signaling can become significant relative to the actual data in the partial bitstream, ultimately reducing coding efficiency. On the other hand, multiplexing may not be suitable for applications that require low delay (for example, for video conferencing applications). With the described multiplexing, the PIPE encoder cannot initiate transmission of a data packet before the partial bit streams have been fully created, since the partition start locations are not known before. Additionally, in general, the PIPE decoder has to wait until it receives the start of the last data segment before it can start decoding a data packet. For applications such as video conferencing systems, these delays can be added to an additional overall system delay of multiple video images (particularly for bit rates that are close to the transmission bit rate and for encoders / decoders that require virtually time interval between two images for encoding / decoding an image), which is critical for such applications. To overcome the disadvantages for certain applications, the PIPE encoder can be configured in a way that the code words that are generated by the two or more binary encoders are interleaved in a single bit stream. The bitstream with the interleaved codewords can be sent directly to the decoder (when a small buffer delay is neglected, see below). On the PIPE decoder side, the two or more binary decoders read the codewords directly from the bitstream in decoding order; decoding can be started with the first bit received. Furthermore, no secondary information is required to signal the multiplexing (or interleaving) of the partial bit streams.

The basic structure of a codeword interleaved PIPE encoder is shown in Fig. 7. Binary encoders 10 do not write the codewords directly into the partial bitstreams, but instead connect to a single memory buffer of codewords 29, from which codewords are written to bitstream 34 in coding order. Binary 10 encoders send requests for one or more codeword buffer 28 entries to codeword buffer 29 and then send codewords 30 to codeword buffer 29, which are stored in reserved buffer entries. Codewords 31 (generally of variable length) from codeword buffer 29 are accessed by a codeword writer 32, which writes the corresponding bits 33 into the produced bitstream 34. Codeword buffer 29 operates as a first-in-first-out buffer; codeword entries that are reserved before are written to the bitstream earlier.

In a further generalization, multiple codeword buffers and partial bit streams 12 are possible, where the number of codeword buffers is less than the number of binary encoders. A binary encoder 10 reserves one or more codewords in the codeword buffer 29, whereby the reservation of one or more codewords in the codeword buffer is triggered by certain events in the memory binary 8 intermediate connected. In one embodiment, the codeword buffer 29 operates in a way that the PIPE decoder can instantly decode the bitstream 34 corresponding to 132 in Fig. 1a and 134 in Fig. 2, respectively. The encoding order in which the codewords are written to the bitstream is the same as the order in which the corresponding codewords are reserved in the codeword buffer. In one embodiment, each binary encoder 10 reserves a codeword, the reservation being triggered by a certain event in the connected binary buffer. In another embodiment, each binary encoder 10 reserves more than one codeword, the reservation being triggered by a certain event in the connected binary buffer. In a further embodiment, the binary encoders 10 reserve a different number of code words, where the amount of code words that is reserved by a particular binary encoder may depend on the particular binary encoder and / or other properties of the encoder. binary / particular binary buffer (such as the associated probability measure, the number of bits already written, etc.).

In one embodiment, the codeword buffer functions as follows. If a new binary 7 is sent to a particular binary 8 buffer and the number of binaries already stored in the binary buffer is zero and no codeword is currently reserved in the codeword buffer for the encoder of binary that is connected to the particular binary buffer, the connected binary encoder 10 sends a request to the codeword buffer, whereby one or more codeword entries are reserved in the word buffer code 29 for the particular binary encoder. Codeword entries can have a variable number of bits; an upper threshold for the number of bits in a buffer entry is usually given by the maximum size of the codeword for the corresponding binary encoder. The next codeword or the following codewords that are produced by the binary encoder (for which codeword input or codeword entries have been reserved) are stored in the reserved input or entries of the codeword buffer. If all reserved buffer entries in the codeword buffer for a particular binary encoder are filled with codewords and the next binary is sent to the binary buffer that is connected to the particular binary encoder, one or more new codewords are reserved in the codeword buffer for the particular binary encoder, etc. Codeword buffer 29 represents a first-in-first-out buffer in a certain way. Entries to the buffer are reserved in sequential order. Codewords for which the corresponding buffer entries have been reserved before are written to the bitstream earlier. Codeword writer 32 checks the status of codeword buffer 29 both continuously and after codeword 30 is written to codeword buffer 29. If the first entry to memory buffer contains a complete codeword (i.e. the buffer entry is not reserved, but includes a codeword), the corresponding codeword 31 and the corresponding buffer entry are removed from the buffer of codewords 20 and codeword 33 bits are written to the bitstream. This process is repeated until the first buffer entry does not contain a codeword (that is, it is reserved or free). At the end of the decoding process, that is, if all source symbols of the considered amount of data have been processed, the codeword buffer should be purged. For that purge process, the following apply to each binary / binary encoder buffer as a first step: if the binary buffer does not contain binaries, a binary with a particular value or arbitrary value is added until the resulting binary sequence represents a binary sequence that is associated with a codeword (as noted above, a preferred way to add binaries is to add those binary values that produce the shortest possible codeword - or one of them - which is associated with a binary sequence containing the original content of the binary buffer as a prefix), then the codeword is written to the next reserved buffer entry for the corresponding binary encoder and the (corresponding) binary buffer. If more than one buffer entry has been reserved for one or more binary encoders, the codeword buffer may still contain reserved codeword entries. In that case, these word inputs from code are populated with arbitrary, but valid, codewords for the corresponding binary encoders. In a preferred embodiment, the shorter valid code word or one of the shorter valid code words (if there are multiple) is inserted. Finally, all the remaining codewords in the codeword buffer are written to the bitstream.

Two examples for the codeword buffer status are illustrated in Fig. 8. In example (a), the codeword buffer contains 2 entries that are populated with a codeword and 5 reserved entries. In addition, the next entry in the free buffer is marked. The first entry is populated with a code word (that is, binary encoder 2 only writes one code word to a previously reserved entry). In the next stage, this codeword will be removed from the codeword buffer and written to the bitstream. Then the first codeword reserved for binary encoder 3 is the first buffer entry, but this entry cannot be removed from the codeword buffer, since it is only reserved, but no input has been written code word in this entry. In example (b), the codeword buffer contains 3 entries that are populated with a codeword and 4 reserved entries. The first entry is marked as reserved and therefore the codeword writer cannot write a codeword to the bit stream. Although 3 codewords are contained in the codeword buffer, the codeword writer has to wait until the codeword is written to the first buffer entry reserved for binary encoder 3. Note that codewords must be written in the order in which they were reserved. To be able to invert the process on the decoder side (see below).

The basic structure of a codeword interleaving PIPE decoder is shown in Fig. 9. The binary decoders 10 do not read the codewords directly from separate partial bit streams, but instead connect to a bit buffer 38 , from which the code words 37 are read in the coding order. It should be noted that bit buffer 38 is not necessarily required, since codewords could be read directly from the bit stream. Bit buffer 38 is mainly included in the illustration to clearly separate different aspects of the processing chain. Bits 39 of bitstream 40 with interleaved codewords, which therefore correspond to bitstream 234 in FIG. 2, are inserted sequentially into bit buffer 38, which represents a first-to-first-enter buffer. go out. If a particular binary decoder 22 receives a request for one or more binary streams 35, binary decoder 22 reads one or more code words 37 from bit buffer 38 through bit requests 36. The decoder PIPE can instantly decode source symbols. Note that the PIPE encoder (as described above) must ensure by proper operation of the codeword buffer that codewords are written in the same order to the bitstream in which they are requested by binary decoders. In the PIPE decoder, the entire decoding process is triggered by requests for source symbols. Parameters such as the number of codewords that are reserved on the encoder side by a particular binary encoder and the number of codewords that are read by the corresponding binary decoder must be the same.

In a further generalization, multiple codeword buffers and partial bitstreams are possible, where the number of bit buffers is less than the number of binary decoders. A binary decoder 22 reads one or more codewords from bit buffer 38 in an instant of time, whereby the reading of the one or more codewords from bit buffer is triggered by certain events in the connected binary buffer 20. In one embodiment, the decoder is operated in a way that one or more codewords are read when a request for a binary 19 is sent to a particular binary buffer 20 and the binary buffer does not contain any binary. But it is also possible to trigger the reading of the codewords by other events, for example if the number of binaries in the binary buffer is below a predefined threshold. In one embodiment, each binary decoder 22 reads a code word, reading being triggered by a certain event in the connected binary buffer. In another embodiment, each binary decoder 22 reads more than one codeword, reading being triggered by a certain event in the connected binary buffer. In a further embodiment, the binary decoders 22 read a different number of code words, where the number of code words that are read by a particular binary decoder may depend on the particular binary decoder and / or other properties of the decoder. binary / particular binary buffer (such as the associated probability measure, the number of bits already read, etc.).

In one embodiment, reading the codewords from the bit buffer works as follows. If a new binary request 19 is sent from the binary buffer selector 18 to a particular binary buffer 20 and the number of binaries in the binary buffer is zero, the connected binary decoder 22 reads one or more code words 37 from bit buffer 38, via request bit 36 to bit buffer 38. Binary decoder 22 converts read code words 37 into binary sequences 21 and stores these sequences of binary in memory connected binary buffer 20. As a final response to the request for a binary 19, the first inserted binary is removed from the binary buffer 20 and sent to the binary buffer selector 18. In response to additional binary requests, the remaining binaries in the binary buffer are removed until the binary buffer is flushed. An additional binary request activates the binary decoder to read one or more new codewords from the bit buffer, etc. Bit buffer 38 represents a first-in-first-out buffer of a predefined size and is continually filled with bits 39 from bitstream 40. To ensure that codewords are written to the bitstream in it As requested by the decoding process, the codeword buffer on the encoder side can function as described above.

Thus, each of the plurality of entropic decoders may be a variable-length decoder configured to map fixed-length codewords into variable-length symbol sequences, and a codeword input such as the output of codeword buffer 43 for receiving a single stream of interleaved codewords. The plurality of entropic decoders 22 can be configured to retrieve the codewords from the codeword input in a sequential order depending on an order in which the symbols are retrieved from the symbol sequence to be reconstructed by selector 18 of the plurality of entropic decoders resulting in a new sequence of symbols to be mapped from a new code word to the respective entropic decoders.

Variable length code word interleaving with low delay restriction

The codeword interleaving described for PIPE encoding does not require that any partition information be sent as secondary information. And since the codewords are embedded in the bitstream, the delay is generally small. However, there is no guarantee that a particular delay constraint will be obeyed (for example specified by a maximum number of bits that are stored in the codeword buffer). Additionally, the buffer size required for codeword buffering can theoretically be made very large. When considering the example in Fig. 8 (b), it would be possible that no additional binaries are sent to the binary buffer 3 and therefore the binary encoder 3 will not send any new codeword to the word buffer code until the purge process is applied to the end of the data packet. So all the codewords for binary encoders 1 and 2 would have to wait until the end of the data packet, before they can be written to the bitstream. This drawback can be overcome by adding an additional mechanism to the PIPE encoding process (and also to the PIPE decoding process as described below). The basic concept of this additional mechanism is that if a measure in relation to the delay or an upper limit of the delay (see below) exceeds a specified threshold, the first entry in the reserved buffer is filled by purging the buffer from corresponding binary (using a mechanism similar to the one at the end of the data packet). Using such a mechanism, the number of standby buffer entries is reduced until the associated delay measure is less than the specified threshold. On the decoder side, binaries that have been inserted on the encoder side to obey the delay reception should be discarded. Basically the same mechanism can be used for this discard of the binaries as on the encoder side. Two embodiments for such delay control are described below.

In one embodiment, the measure for delay (or an upper bound on the delay) is the number of active buffer entries in the codeword buffer, where the number of active buffer entries is the number of input entries. reserved buffer plus the number of buffer entries that contain codewords. Note that the first buffer entry is always a reserved buffer entry or a free buffer entry, since if the first buffer entry contains a codeword, this codeword is written to the bitstream . If, for example, the maximum allowed buffer delay (as determined by the application) is D bits and the maximum codeword size for all binary encoders is L, a lower bound for the maximum number of words Code that can be contained in the codeword buffer without violating the delay constraint can be calculated by N = D / L. The mean bit delay D is not required by the system, but the maximum number of codewords N must be known for both the encoder and the decoder. In one embodiment, the maximum number of codeword buffer entries N is set by the application. In another embodiment, the maximum number of N- word buffer entries is signaled inside the bitstream, for example, in the data packet header (or fragment header) or in a set of parameters, which is included in a bit stream. If a binary encoder 10 sends a request to reserve one or more new buffer entries to codeword buffer 29, the following process is executed before a new buffer word entry is reserved. code (i.e. run multiple times if multiple codeword buffer entries are reserved by request): if the number of currently active buffer entries plus 1 (taking into account the buffer entry to be reserved below) is greater than the maximum number of buffer entries in Codewords N, the first buffer entry (which is reserved) is purged by the process described below until the number of currently active buffer entries plus 1 is less than or equal to the maximum number of buffer entries N codeword. Purging a reserved buffer entry is similar to purging at the end of a data packet. The binary encoder 10 that has reserved the first corresponding buffer entry is purged by adding binaries with particular or arbitrary values to the connected binary buffer 8 until the resulting binary sequence represents a binary sequence that is associated With a codeword, the codeword is then written to the reserved buffer entry and is finally added to the bit stream (while clearing the binary buffer and removing the previously reserved buffer entry). As mentioned above, a preferred way to add binaries to the binary buffer is to add those binaries that produce the shortest possible codeword. On the decoder side, a similar process is run to discard binaries that have been added to obey the delay constraint. Therefore, the decoder maintains a counter C that counts the codewords that have been read from the bit buffer (this counter can be kept in the bit buffer). Counter C is initialized (eg zero) at the start of decoding a data packet incremented by one after a codeword is read. Furthermore, each binary decoder 22 contains a Cx counter, which stores the value of the C codeword counter before the last codeword is read by the corresponding binary decoder 22, i.e. when a binary decoder 22 home reads a new codeword, its counter Cx is set equal to C as a first stage, and then the codeword is read from the bit buffer. When a request for a binary 19 is sent to a particular binary buffer 20 and the difference (C-Cx) between the global code word counter C and the counter Cx of the connected binary decoder 22 is greater than the maximum number of codeword buffer entries N, all binaries currently stored in the particular binary buffer 20 are discarded and ignored. Along with that additional stage, decoding is operated as described above. If the binary buffer 20 to which a request for a binary 19 is sent is empty (either because all binaries have already been removed or because a low delay mechanism discarded all binaries in the first stage after it has been received the binary request), the connected binary decoder 22 reads one or more new codewords from the bit buffer 38, etc.

In another embodiment, the measure of delay (or an upper bound on delay) is the sum of the maximum codeword lengths for the active buffer entries in the codeword buffer, where the maximum codeword length of Code for a particular buffer entry depends on the decoded binary that is associated with that buffer entry. As an illustration, the maximum codeword lengths for the buffer entries are indicated in the examples in 6. Note again that the first buffer entry is always a reserved buffer entry or a free buffer entry, since if the first buffer entry contains a codeword, this codeword is written to the bitstream. Let the maximum buffer delay allowed (as determined by the application) be D bits. This maximum buffer delay D must be known for both the encoder and the decoder. In a preferred embodiment, the maximum buffer delay D is set by the application. In another preferred embodiment, the maximum buffer delay D is signaled within the bitstream, for example, in the data packet header (or fragment header) or in a parameter set, which is included in the bit stream. It can be signaled in units of bits, or bytes, or a multiple of bits, or a multiple of bytes. If a binary encoder 10 sends a request to reserve one or more new buffer entries to codeword buffer 29, the following process is executed before a new buffer word entry is reserved. code (that is, it runs multiple times if multiple codeword buffers are reserved by request).

If the sum of the maximum codeword lengths for all currently active buffer entries plus the maximum codeword length for the reserved buffer entry is greater than the maximum buffer delay D, the first buffer entry (which is reserved) is purged by the process described above until the sum of the maximum codeword lengths for all active buffer entries plus the maximum codeword length for the buffer entry to be reserved is less than or equal to the maximum buffer delay D. As an example, consider the example in Fig. 8 (b). The sum of the maximum codeword lengths for all currently active buffer entries is 29. Suppose the maximum buffer delay D is set equal to 32. If the next buffer entry is reserved by the encoder binary 2 for which the maximum codeword length equals 3, the first buffer entry is not purged, since 29 3 is not greater than 32. But if the next buffer entry is reserved by the binary encoder 1 for which the maximum codeword length equals 7, the first buffer entry is purged, since 29 7 is greater than 32. Purging of the reserved buffer entry is performed as it has been described previously (by adding binary with particular or arbitrary values from the corresponding binary buffer).

On the decoder side, a similar process is run to discard binaries that have already been added to obey the delay constraint. Therefore, the decoder maintains a counter C that counts the maximum codeword length for codewords that have been read from the bit buffer (this counter can be kept in the bit buffer). Note that the maximum codeword lengths that are associated with different binary decoders may be different. Counter C is initialized (for example zero) at the start of decoding a data packet and incremented after a codeword is read. This counter is not incremented by the actual length of the codewords read, but by its maximum length, i.e. if a codeword is read by a particular binary decoder and the maximum codeword length that is associated with the codeword table used by the particular binary decoder is Lx (a different binary decoder can be associated with a different maximum codeword length), the counter C is incremented by Lx. In addition to the global counter C, each binary decoder 22 contains a counter Cx, which stores the value of the codeword counter C before the last codeword is read by the corresponding binary decoder 22, i.e. when a Particular binary decoder 22 reads a new codeword, its counter Cx is set equal to C as a first stage, and then the codeword is read from the bit buffer. When a request for a binary 19 is sent to a particular binary buffer 20 and the difference (C-Cx) between the global counter C and the counter Cx of the connected binary decoder 22 is greater than the maximum buffer delay D , all binaries that are currently stored in the particular binary 20 buffer are discarded and ignored. Along with that additional stage, decoding works as described above. If the binary buffer 20 to which the request for a binary 19 is sent is empty (because all binaries have already been removed or because the low delay mechanism discarded all binaries in the first stage after it has been received the binary request), the connected binary decoder 22 reads one or more new codewords from bit buffer 38, etc.

Furthermore, the plurality of entropic decoders 22 and selector 18 can be configured to intermittently discard symbol sequence suffixes so that they do not participate in the formation of the symbol sequence to be reconstructed 29. Intermittent discard can be performed in events where a The number of codewords that have been retrieved from the codeword entry by the plurality of entropic decoders between two consecutive codeword recoveries of a respective entropic decoder from the codeword entry meets a predetermined criterion. The plurality of entropy encoders and the codeword buffer may, in turn, be configured to intermittently extend currently submitted but not yet mapped symbols to valid symbol sequences without regard to currently submitted but not yet mapped symbols as prefix, map the symbol sequences thus extended into codewords, enter the codewords thus obtained in the reserved codeword entries and purge the codeword entries. Intermittent extension, input and purging can take place with events where a number of reserved codeword entries plus a number of codeword entries having codewords entered therein meets a predetermined criterion. The predetermined criterion may take into account the maximum codeword lengths of the plurality of encoder / decoder pairs.

For some architectures, the above described preferred embodiment for interleaving codewords could result in a drawback in terms of decoding complexity. As illustrated in Fig. 9, all binary decoders 22 read codewords (in the general case, variable length codewords) from a single bit buffer 38. Reading the codewords cannot be carried out in parallel, since the code words must be read in the correct order. This means that a particular binary decoder must wait until other binary decoders finish reading the codewords. And when the complexity of reading the variable length codewords is significant relative to the rest of the decoding processes (partially parallel), this variable length codeword access can be a bottleneck for everything the decoding process. There are some variations of the written embodiments that can be employed to reduce the complexity of access from the single bit buffer, a few of which will be described below. In a preferred embodiment, there is a single set of code words (representing for example a non-redundant prefix code) and the set of code words that is used for each binary decoder 22 is a subset of a single set of words code. Note that different binary decoders 22 can use different subsets of the single set of code words. Even if the codeword sets that are used by some of the binary decoders 22 are the same, their association with the binary sequences is different for different binary decoders 22. In a particular embodiment, the same set of code words code is used for all binary decoders 22. If you have a single codeword set that includes codeword sets for all binary decoders as subsets, parsing of codewords can be done outside of binary decoders, which can reduce the complexity of accessing the codeword. The PIPE encoding process is not changed relative to the previously described process. The modified PIPE decoding process is illustrated in Fig. 10. A single codeword reader is fed with bits 46 from bit stream 40 and parses codewords - generally of variable length -. The code words 44 read are inserted into a code word buffer 43, which represents a first-in-first-out buffer. A binary decoder 22 sends a request for one or more codewords 41 to the codeword buffer 43 and in response to this request, the one or more codewords are removed from the codeword buffer ( in sequential order) and sent to the corresponding binary decoder 22. Note that with this embodiment, the analysis of potentially complex code words can be performed in a background process and does not need to wait for binary decoders. Binary decoders access code words already parsed, parsing of potentially complex code words is not part of a request to the global buffer. Instead the already parsed codewords are sent to the binary decoders, which can also be implemented in a way that only the codeword indexes are sent to the binary decoders.

Fixed length bitstream interleaving

A further way to reduce the complexity of the PIPE decoder can be achieved when the binary decoders 22 do not read variable length codewords from the global bit buffer 38, but instead always read fixed length bit streams from the memory. global bit buffer 38 and add these fixed-length bit streams to a local bit buffer, where each binary decoder 22 connects to a separate local bit buffer. Variable length codewords are then read from the local bit buffer. Therefore, the analysis of the variable length codewords can be done in parallel, only the access of fixed length bitstreams has to be done in a synchronized way, but such access of fixed length bitstreams is usually very fast , so that the overall decoding complexity can be reduced for some architectures. The fixed number of binaries that are sent to a particular local bit buffer may be different for different local bit buffers and may also vary over time, depending on certain parameters such as events in the binary decoder, buffer binary, or bit buffer. However, the number of bits read on a particular access does not depend on the actual bits that were read during the particular access, which is the important difference from reading variable length codewords. The reading of fixed length bitstreams is triggered by certain events in the binary buffers, binary decoders, or local bit buffers. As an example, it is possible to request the reading of a new fixed-length bit stream when the number of bits that are present in a connected bit buffer falls below a predefined threshold, where different threshold values can be used for different bit buffers. In the encoder, it must be ensured that the binary fixed-length sequences are inserted in the same order in the bitstream, in which they are read from the bitstream on the decoder side. It is also possible to combine this interleaving of fixed length sequences with a low delay control similar to that explained above. Next, a preferred embodiment for interleaving of fixed-length bit sequences is described.

Fig. 11 shows an embodiment of a PIPE encoder structure for the embodiment that interleaves fixed-length bit sequences for two or more binary encoders. Unlike the embodiment depicted in Fig. 7, binary encoders 10 do not connect to a single codeword buffer. Instead, each binary encoder 10 is connected to a separate bit buffer 48, which stores bits for the corresponding partial bitstream. All bit buffers 48 are connected to a global bit buffer 51. Global bit buffer 51 is connected to a bit writer 53, which deletes bits 52 in the encoding / decoding order from the buffer global bitstream and writes the deleted bits 54 into bitstream 55. In a certain event in a particular bit buffer 48 or the connected binary encoder 10 or binary buffer 8, the bit buffer 48 sends a request 49 to the global bit buffer 51 by which a certain number of bits are reserved in the global bit buffer 51. The requests for the reservation of fixed length bitstream 49 are processed in sequential order. The global bit buffer 51 represents a first-in-first-out buffer in a certain way; bits that are reserved before are written before in the bitstream. It should be noted that different bit buffers 48 may reserve a different number of bits, which may also vary over time based on already encoded symbols; but the number of bits that are reserved for a particular request is known at the time the request is sent to the global bit buffer.

In one particular, the bit buffer 48 and the global bit buffer 51 are operated as described below. The number of bits that is reserved by a particular bit buffer 48 is indicated by Nx. This number of bits Nx may be different for different bit buffers 48 and may also vary over time. In one embodiment, the number of bits Nx that is reserved for a particular bit buffer 48 is fixed over time. Reservations for a fixed number Nx of bits 49 are triggered based on the number of bits Mx in bit buffer 48, the number of bits Nx for reservation requests, and the associated maximum code word length Lx . Note that each binary encoder 10 can be associated with a different maximum codeword length Lx . If a binary 7 is sent to a particular binary 8 buffer, and the particular binary 8 buffer is empty, and no more than one Nx bit sequence is reserved in the global bit buffer for bit buffer 48 that connects to the memory particular binary buffer (through a binary encoder), and the difference Nx - Mx between the number Nx of bits that are reserved by a reservation request from the connected bit buffer 48 (through an encoder binary) with the particular binary buffer 8 and the number of bits Mx that is currently present in this bit buffer 48 is less than the maximum codeword length Lx that is associated with the corresponding binary encoder 10, the connected bit buffer 49 sends a request 49 for the Nx bit reservation to the global bit buffer 51. The global bit buffer 51 reserves Nx bits for the m bit buffer 48 particular and increments its pointer for the next reservation. After the Nx bits have been reserved in the global bit buffer, binary 7 is stored in binary buffer 8. If this single binary already represents a sequence of binaries that is associated with a codeword, the binary encoder 10 removes this binary from binary buffer 8 and writes the corresponding code word 47 to the connected bit buffer 48. Otherwise (this single binary already represents in a sequence of binaries that is associated with a codeword), additional binaries 7 are accepted by the particular binary 8 buffer until the binary 8 buffer contains a binary sequence which is associated with a code word. In this case, the connected binary encoder 10 removes the binary sequence 9 from the binary buffer 8 and writes the corresponding code word 47 to the connected bit buffer 48. If the resulting number of bits Mx in the bit buffer 48 is greater than or equal to the number of bits Nx reserved, the Nx bits that were written first in the bit buffer 48 are inserted into the space previously reserved in the global bit buffer 51. For the next binary 7 that is sent to the particular binary 8 buffer, the same process specified above is executed; that is, it checks first if a new number of Nx bits should be reserved in the global bit buffer (if Nx - Mx is less than Lx) and then the binary is inserted into binary buffer 8, etc. The bit writer writes the fixed-length bit streams of the global bit buffer in the order in which they were reserved. If the first fixed-length entry in the global bit buffer 51 contains a fixed-length bit stream that has actually been inserted into the global bit buffer (that is, it is not just reserved), the bit writer 53 deletes the bits in this bit stream 52 from the global bit buffer 51 and writes the bits 54 to the bit stream. This process is repeated until the first fixed-length entry in the global bit buffer represents either a reserved or a free entry. If the first fixed-length entry in the global bit buffer represents a reserved entry, bit writer 53 waits until this entry is filled with the actual bits before writing additional bits 54 in bit stream 55.

At the end of a data packet, the binary buffers are purged as previously described. Also, bit buffers must be purged by adding bits with a particular or arbitrary value until all reserved buffer entries in the global bit buffer are filled and written to the bitstream.

Two examples of possible states of global bit buffer 51 are illustrated in Fig. 12. In example (a), a case is illustrated where different bit buffers / binary encoders reserve a different number of bits . Global bit buffers contain 3 inputs with actually written fixed-length bitstreams and 4 inputs with reserved fixed-length bitstreams. The first fixed-length entry already contains actual bits (which must have been recently inserted by bit buffer / binary encoder 2); This input (i.e. the corresponding 8 bits) can be removed and written to the bit stream. The following entry reserves 10 bits for binary encoder 3, but the actual bits have not been inserted yet. This input cannot be written to the bit stream; it must wait until the actual bits are inserted; in the second example (b), all bit buffers / binary encoders reserved the same number of bits (8 bits). The global bit buffer contains 4 reserves for 8 bit streams and 3 actually written 8 bit streams. The first entry contains an 8-bit reservation for binary encoder 3. Before new bits can be written to the bitstream, the bit writer has to wait until the bit buffer / binary encoder 3 writes the values of the 8 bits in the reserved input.

Fig. 13 shows an illustration of a PIPE decoder structure for the embodiment of the invention that interleaves fixed-length bit sequences. Unlike the embodiment shown in Fig. 9, the binary decoders 22 are not connected with a single bit buffer. Instead, each binary decoder 22 is connected to a separate bit buffer 58, which stores the bits from the corresponding partial bitstream. All bit buffers 58 are connected to a global bit buffer 61. Bits 62 from bit stream 63 are inserted into global bit buffer 61. In certain events, in a particular bit buffer 58 or connected binary encoder 22 or binary buffer 20, bit buffer 58 sends a request 59 to global bit buffer 61 by which a fixed length bit stream 60 is removed from the bit buffer global 61 and inserted into the particular bit buffer 58. Requests for fixed length bit streams 59 are processed in sequential order. The global bit buffer 61 represents a first-in-first-out buffer; bits that are inserted earlier into the global bit buffer are removed earlier. It should be noted that different bit buffers 58 may request a different number of bits, which may also vary over time based on already decoded symbols; but the number of bits that is requested by a particular request is known at the time the request is sent to the global bit buffer. It should be noted that the global bit buffer 61 is not actually required, since the codewords could also be read directly from the bit stream. Global bit buffer 61 is mainly included in the illustration to clearly separate different aspects of the processing chain.

In one embodiment, bit buffers 58 and global bit buffer 61 are operated as described below. The number of bits that are requested and read by a particular bit buffer 58 is indicated as Nx, it is equal to the number of bits that are written to the global bit buffer by the corresponding bit buffer on the encoder. This number of bits Nx may be different for different bit buffers 58 and may also vary over time. In a preferred embodiment of the invention, the number of Nx bits that are requested and read by a particular bit buffer 58 is fixed over time. The reading of a fixed number Nx of bits 60 is triggered based on the number of bits Mx in bit buffer 58 and the maximum length of the associated code word Lx . Note that each binary decoder 22 can be associated with a different maximum codeword length Lx . If a request for a binary 19 is sent to a particular binary 20 buffer, and the particular binary 20 buffer is empty, and the number Mx of bits in the connecting bit buffer 58 (via a binary decoder) with the particular binary buffer 20 being less than the maximum codeword length Lx that is associated with the corresponding binary decoder 22, the connected bit buffer 58 sends a request 59 for new streams Nx bits to global bit buffer 61. In response to this request, the first Nx bits are removed from global bit buffer 61 and this sequence of Nx bits 60 is sent to bit buffer 58 from which the request was sent. Finally, this Nx bit sequence is added to the corresponding bit buffer 58. The next codeword 57 is then read from this bit buffer, and the connected binary decoder 22 inserts the associated binary sequence 21 into the connected binary buffer 20. As a final response to the original request for a binary 19, the first binary is removed from the binary buffer 20 and this decoded binary 25 is sent to the binary buffer selector 18. When the next binary request 19 is sent to the particular binary 20 buffer and memory binary buffer is not empty, the next bit is removed from the binary buffer 20. If the binary buffer is empty but the Mx number of bits in the connected bit buffer 58 is greater than or equal to the length maximum associated Lx codeword, the next codeword is read from the bit buffer and a new sequence of binaries is inserted into the memory binary buffer, from which the first bit is cleared and sent to the binary buffer selector. If the binary buffer is empty and the Mx number of bits in the connected bit buffer 58 is less than the maximum length of the connected Lx codeword, the following sequence of Nx bits is read from the bit buffer global 61 and inserted into the connected local bit buffer 58, the next codeword is read from the bit buffer, a new binary stream is inserted into the binary buffer, and the first binary in the stream it is removed and sent to the binary buffer selector. This process is repeated until all the source symbols are decoded.

At the end of a data packet, more binary and / or bits than required for decoding the required source symbols could be inserted into the binary buffer and / or bit buffer. The remaining binaries in the binary buffer and the remaining bits in the bit buffer are discarded and ignored.

Fixed-length bitstream interleaving with low delay constraint

The embodiment described for a PIPE encoder and decoder with fixed-length bitstream interleaving can also be combined with the scheme for encoder buffer delay control, which has been described above, the concept of PIPE encoding is the same as in the embodiment with the delay control described above. If a delay-related measure or upper delay delimitation (see below) exceeds a specified threshold, the first reserved buffer entry is populated by purging the corresponding binary buffer (using a similar mechanism as in end of a data packet) and potentially writing additional bits to fill all bits of the reserved fixed-length buffer entry. Using such a mechanism, the number of standby buffer entries is reduced until the associated delay measure is less than the specified threshold. On the decoder side, binaries and bits that have been inserted on the encoder side in order to obey the delay constraint must be discarded. Basically the same mechanism can be used for this discard of binaries and bits as on the encoder side.

In one embodiment, the measure for delay (or an upper bound on delay) is the number of bits in the active buffer entries in the global bit buffer, where the number of active buffer entries is the number of reserved fixed-length buffer entries plus the number of fixed-length buffer entries that contain bits already written. Note that the first buffer entry is always a reserved fixed-length buffer entry or a free buffer entry, since if the first buffer entry contains written bits, these bits are written to the bitstream. Let the maximum allowable buffer delay (as determined by the application) be D bits. This maximum buffer delay D must be known for both the encoder and the decoder. In a preferred embodiment of the invention, the maximum buffer delay D is set by the application. In another preferred embodiment of the invention, the maximum buffer delay D is signaled within the bitstream, for example in the packet data header (or fragment header) or in a parameter set, which is included in the bit stream. It can be signaled in units of bits, or bytes, or a multiple of bits, or a multiple of bytes. If a binary encoder 10 sends a request for the reservation of a new fixed length bit stream to the global bit buffer 51, the following process is executed before a new entry is reserved in the fixed length buffer .

If the number of bits in the active buffer entries in the global bit buffer plus the number of bits to be reserved by the current reservation request is greater than the maximum buffer delay D, the first buffer entry (which is reserved) is purged by the process described below until the number of bits in the active buffer entries in the global bit buffer plus the number of bits to be reserved by the current reservation request is less than or equal to the maximum buffer delay D. Purging a reserved fixed-length buffer entry is similar to purging at the end of a data packet. The binary encoder 10 that connects to the bit buffer 48 that has reserved the first corresponding buffer entry is purged by adding binaries with particular or arbitrary values to the connected binary buffer 8 until the sequences of Resulting binary represents a binary sequence that is associated with a codeword, the codeword is then inserted into the corresponding bit buffer 48. As mentioned above, a preferred way to add binaries to the binary buffer is to add those binaries that produce the shortest possible codeword. If, after writing the codeword to the connected bit buffer and the potential insertion of a fixed-length bit stream into the global bit buffer, there are still bits in the bit buffer (it is (codeword writing does not completely fill the reserved fixed-length bitstream), additional bits with particular or arbitrary values are added to the bit buffer until all bits are removed from the bit buffer and are written to the reserved buffer entry. Finally, at the end of this process, the completed buffer entry (the first fixed-length entry in the global bit buffer) is removed from the global bit buffer and written to the bit stream.

On the decoder side, a similar process is run to discard the binaries and bits that have been added to obey the delay constraint. Therefore, the decoder maintains a counter C that counts the bits that have been read from the global bit buffer (this counter can be kept in the global bit buffer). Counter C is initialized (eg zero) at the start of decoding a data packet and incremented after a fixed length sequence is read. If a fixed length Nx bit sequence is read from the global bit buffer 61, the counter C is incremented by Nx. In addition to the global counter C, each bit buffer 58 contains a counter Cx, which stores the value of the bit counter C before the last fixed-length bit stream is read within the corresponding bit buffer 58. When a particular bit buffer 58 reads a new fixed-length bit stream, its counter Cx is set equal to C as a first step, and then the fixed-length bit stream is read from the global bit buffer 61. When a request for a binary 19 is sent to a particular binary buffer 20 and the difference (C-Cx) between the global counter C and the counter Cx of the connected bit buffer 58 is greater than the maximum memory delay Buffer D, all binaries that are currently stored in the particular binary buffer 20 and all bits that are stored in the connected bit buffer 58 are discarded and ignored. Along with that additional stage, decoding is operated as described above. If the binary buffer 20 to which a request for a binary 19 is sent is empty (either because all binaries have already been removed or because a low delay mechanism discarded all binaries in the first stage after has received the binary request), the connected binary decoder 22 tries to read a new codeword from the connected bit buffer 58. If the number of bits in bit buffer 58 is less than the maximum codeword length, a new fixed-length bitstream is read from global bit buffer 61, before the buffer word is read. code etc.

The embodiments which have been described with respect to Figs. 7-13 refer to possibilities for achieving a bitstream path sandwiched between PIPE encoder 104 on the one hand and PIPE decoder 202 on the other hand. As described above with respect to Figs. 1 and 2, the encoding and decoding apparatus Entropic can be connected to each other by two separate channels, one of which carries the VLC 112 bitstream and the other of which carries the interleaved PIPE encoded bitstream. However, there are also possibilities for interleaving even both VLC bit streams 112 as well as PIPE 118 encoded bit streams, and those possibilities will be described below with respect to Figs. 20-24. However, prior to that, a mathematical basis regarding the PIPE coding scheme is provided as well as details regarding how to optimally subdivide the range of probabilities by assigning the resulting individual partial results to the encoders. 116 entropics and 210 individual entropic decoders, respectively.

As already indicated, in PIPE encoding the uniform space of the discrete symbol input sequence is mapped over a small set of binary probability intervals. The probability models for the source symbols can be fixed or adaptive while the entropic coding using the probability intervals remains fixed and is decoupled from the modeling stage. Each of the probability ranges can be coded using a very simple entropy code that has the level of complexity of Huffman codes. The excess rate of the partition interval probability entropy code (PIPE) is similar to that of pure arithmetic coding.

Entropic encoding, in general, can be considered as the most generic form of lossless data compression. Lossless data compression is intended to represent discrete data with fewer bits than required for the original data representation but without any loss of information. Discrete data can be in the form of text, graphics, images, video, audio, voice, facsimile, medical data, weather data, financial data, or any other form of digital data. In many encoding applications, the original source data is first mapped into so-called encoding symbols, and these encoding symbols are then encoded entropically. The mapping into encoding symbols may include quantization, in which case the overall encoding scheme is lossy. An encoding symbol s can take any value from an M-Aryan alphabet (M> 2) A = {ao, ..., aM-1}. For the purpose of encoding the symbol s, the alphabet is associated with an estimated probability mass function (pmf) {ps (ao), K, ps (aM-1)} and all dependencies between encoding symbols that are not consider in this pmf they despise themselves. For these abstract settings, entropy

M-1

P ^{s (} a ^¡ ) log ² P ^{s (} a ^¡ )

H ^s = ^{'= °} (B1)

it is the largest lower bound for the expected word length in bits per symbol, for encoding the s symbols, which can be achieved with entropic encoding techniques. For decades, Huffman coding and arithmetic coding have dominated practical entropic coding. There are well-known examples of practical codes capable of approaching the entropic limit (in a certain sense).

For a fixed probability distribution, Huffman codes are relatively easy to construct. The most attractive property of Huffman codes is that they can be implemented efficiently by using variable length code tables (VLC). However, when dealing with time-varying source statistics, i.e. probabilities of changing symbols, the adaptation of the Huffman code and its corresponding VLC tables is quite demanding, both in terms of algorithmic complexity as well as in terms of implementation costs. . Also, in the case of having a dominant alphabet value with P ^s ( a) > 0.5, the redundancy of the corresponding Huffman code (without the use of any alphabet extension such as string length encoding execution) it can be quite substantial. Other drawbacks to Huffman codes come from the fact that in dealing with higher order probability modeling, multiple sets of VLC tables may be required.

Arithmetic coding, on the other hand, although substantially more complex than VLC, offers the advantage of more consistent and appropriate handling when handling adaptive and higher-order probability modeling as well as in the case of highly skewed probability distributions. . Actually, this feature is basically the result of the fact that arithmetic coding provides a mechanism, at least conceptually, to map any given probability estimate value in a more or less direct way to a part of the resulting codeword. Being provided with such an interface, arithmetic coding allows a clear separation between probability modeling and probability estimation tasks, on the one hand, and actual entropic coding, i.e. mapping symbols to codewords, on the other hand.

Unlike the recently explained conventional entropic coding schemes, PIPE coding uses partitioning of probability intervals, the mathematical basis of which is described in more detail below.

Consider the sequence of encoding symbols {so, K s «-1}. Each symbol is drawn from an alphabet s, e A. Los a Q *

alphabets A ⁱ = {0, 1, ...} contain two or more letters each being associated with a probability estimate Ps ( am ). The probability estimates Ps ( am ) are known for encoder and decoder and can be fixed or variable. Variable probabilities are assumed to be estimated simultaneously at the encoder and decoder. Ai alphabets can both be identical for the sequence of symbols and associate different types of symbols with different alphabets. In the latter case, it is assumed that the decoder knows the alphabet of each symbol in the sequence. This assumption is justified since the practical source codec descriptions contain a syntax that stipulates the order of the symbols and their alphabets.

The symbol sequence {s0, K s «-1} is converted to a sequence of binary symbols, which are also called binary. For each symbol to binary conversion sj

b = {b °. } - Yb (s,) (B2)

a 'bi

represents a bijective mapping of letters of the alphabet m into sets of ordered binaries m . Yb binary conversion mapping may be different for different symbols s, or symbol categories. Each sequence of binaries i for a particular symbol s consists of one or more binaries bk. On the decoder side, symbols s can be reconstructed, by inverse mapping s, = '' b ''' given the sequence of binaries b'. As a result of the binary conversion, a binary sequence {bo, K, be-i} is obtained that represents the sequence of source symbols {so, K, s «-i}.

All binary b¡ are associated with the same binary alphabet B = {0,1}, but the corresponding binary pmfs {Po, Pl} being Pl = Po are normally different. A binary pmf {Po, Pl} can be described by the value of the least likely binary ^{b j p j} (LPB) LPB and its probability rLPB (with rLPB <o, 5). This description of probability b j

The binary {LPB, PLPB } can be derived directly from probability estimates Ps (am) for symbol alphabets given the binary conversion maps Yb. It is also possible (and often preferable) to estimate b j pj

directly { LPB , LPB } simultaneously on the encoder and decoder side. Therefore, binaries can be associated with a probability model (also referred to as a context) based on the syntax and symbols or binaries actually encoded. And for each probability model, the description of b j pj

probability { LPB , LPB} can be estimated based on the values of the binaries that are encoded with the probability model. An example of such binary probability modeling is described with respect to H.264 CABAC.

Since the binary entropic function

H ( p) = -p log2 (p) - (1 - p) log2 (1 - p) (B3) is symmetric around p = 0, 5, the same binary encoder can be used for encoding all binaries that they are associated with the same probability LPB pLPB regardless of the value of bj . Therefore the bc bc binary sequence {bo ..., bB-1} becomes a binary encoding sequence { B -1}. For each bi ^Yi nario b, the corresponding bijective mapping is specified , c by

bc¡ = YC ( bj ) = b © b [PB (B4) where © indicates the exclusive OR operator. On the decoder side, bj binaries can be reconstructed given c

^{b bj}

_{encoding binaries} 1 and the corresponding LPB value “LPB by inverse mapping b¡ = ( ^yc )>; ). U ^, n encoded binary ^bC ' = o specifies that the corresponding binary value bj is equal to the value LPB ^b L ^j PB and an encoded binary ^bc 1 = 1 specifies that the value of the corresponding binary bj is equal to the value of the binary bj

most likely (MPB) 1 LPB

The encoded binary sequence { ^b or ^c , K, ^b B ^c 1} only represents the source symbol sequence {so, K, sw-i} and the corresponding probability estimates, which can be used for entropic coding, are

Fully described by the LPB Plpb probabilities ( Plpb being <0.05). Therefore, only the probabilities in the semi-open interval (0, 0.5) need to be considered for the design of the binary entropic encoder c

for encoding binaries b

c

For actual binary entropic encoding, the encoding binary sequence { ^b 0 ^c , K, ^b B-1} is projected over a small number of probability intervals i *. The LPB probability interval (0. 0.5] is partitioned into K intervals Ik = ( pk, pk + 1]

OR ^{ik n ij}

^{= (0 0.5} y ^{= 0 for k t j} (B5)

k = 0

The set of K is characterized by K-1 edges of interval pk with k = 1, K, K-1. Without loss of generality we assume that pk <Pk + 1 for k = 0, K, K. The outer interval edges are fixed and given by p0 = 0 and pk = 0.5. A simple non-adaptive binary entropic encoder is designated for each Ik interval . All bc binaries

Encoding 1 with associated LPB probabilities pLPB and Ik are assigned to the interval Ik and encoded with the corresponding fixed entropic encoder.

In the description that follows, all bC binaries represent encoding binaries and all probabilities p

are LPB pLPB probabilities

To investigate the impact of discretization on the probability interval on coding efficiency, we assume that we can design an optimal entropic encoder for a fixed probability that reaches entropic delimitation. Each probability interval Ik = (pk, pk + 1] is associated with a representative probability pIk and Ik and the corresponding optimal entropic encoder should achieve the entropy limit for its representative probability. Under this assumption, the encoding rate of a binary with probability p using the optimal entropy encoder for the representative interval pIk is given by

R ( p, pk) = -p log2 (pIk) - (1 - p) log2 (1 - pk)

= H ( pIk) + (p - pk) H ' ( pk) (B6)

where H (p) represents the binary entropy function 3 and

it is its first derivative. We further assume that the probability distribution on the interval (0, 0.5)

0.5

í f ( p) dp

is given by / (p), where ⁰ = 1. Then the expected rate, in bits per binary, for a given set of K intervals {Ik} with corresponding representative probabilities {pIk} can be described as

K-1Í pk * 1

É jR (p, p¡k) f ( p) dp

R = R ({Ik}, {pIk}) = k = ° ^ pk (B8)

The first partial derivative with respect to any representative probability pIk, with k = 0, K, K-1, is given by

d

R

= 0 tiene una solución única The equation

= 0 has a unique solution

f Pk +1 pf ( p) dp

p. * _p + 1 f ( p) dp

^{k pk} (B10)

for the representative probability pik within the definition domain lk. The second partial derivative for this solution

is always greater than zero if

f 1 pf ( p) dp

^pk > 0 (B12)

Therefore, if condition B12 is satisfied, the value lk given by equation B10 is the representative probability for an interval lk that minimizes the expected global rate R given the boundaries of the interval pk and pk + 1. Otherwise, no binary is projected in the interval lk and the representative probability pik and lk can be chosen arbitrarily without any impact on the overall rate R; but such configuration should be avoided, since the interval lk should not be used for entropic encoding.

To find a condition for the optimal interval edges, we investigated the first derivatives of the expected global rate R with respect to the interval edges pk with k = 1, K, K-1. If / (p)> 0 for all pe plk-1, plk), the

d _R

equation spk = 0 has a single solution.

H ( pk ) - p, H 'p ) - Hp J pk -HHJ

pk H 'p -,) - H ' ( pk ) (B13)

for the interval edge pk within the definition domain [p «-1, plk) and the second partial derivative for this solution

d2

R ( pk )

dp2

f ( pl) {H '( P ik J - H ' ( P ik )) (B14)

is always greater than zero, so plk is the edge of the interval pk and pk - 1, plk) that minimizes the expected overall rate R given the representative intervals p ^ -1 and plk . If there are probabilities eg plk-1, plk) with fp) = 0, the

^d R

equation dpk = 0 has multiple solutions, but pl as given by equation B13 is still optimal even though additional optimal solutions may exist.

Given the number of intervals K and the probability of distribution / (p), the edges of interval pk, with k = 1, K, K-1, and the representations of the interval plk, with k = 0, K, K-1 , that minimize the expected global rate R can be obtained by solving the system of equations given by equations B10 and B13 subject to the conditions

B12 for k = 0, K, K-1. This can be accomplished with the following iterative algorithm.

Algorithm 1:

1) Partition the interval (0, 0.5] into K arbitrary intervals lk = (pk, pk + 1] with po = 0, pk = 0.5, and p k = 0, K, K-1 such that conditions B12 are obeyed for all k = 0, K, K-1.

2) Update plk representations with k = 0, K, K-1 according to equation B10

3) Update the interval edges pk with k = 1, K, K-1 according to equation B13

4) Repeat the two previous stages until convergence

Figure 14 shows an example for discretizing the optimal range using the algorithm described. For this one For example, we assume a uniform probability distribution f ( p) = 2 for 0 <p <0.5 and the probability interval (0, 0.5] is partitioned into K = 4 intervals. It can be seen that the discretization of the interval of Probability leads to a linear approximation in the form of pieces A (p) of the binary entropy function H (p) where A (p)> H (p) for all P e (0, 0.5].

As a measure for the impact of interval discretization on coding efficiency, the expected overall rate of increase relative to the entropy limit can be used.

R

0.5 - 1

_ [ H (p) f (p) dp

P = (B15)

es igual a 1/(2ln2) bit por binario y la sobrecarga de la tasa p es igual al 1,01 %. La Tabla 4 lista las sobrecargas de tasa Puni y Plin para la distribución de probabilidad uniforme y una distribución de probabilidad de incremento lineal fp)=8p con p e (0, 0,5], respectivamente, para números seleccionados de K intervalos.For the example in Figure 14, the expected value of entropy

equals 1 / (2ln2) bit per binary and the rate overhead p equals 1.01%. Table 4 lists the Puni and Plin rate overloads for the uniform probability distribution and a linear increase probability distribution fp) = 8p with p e (0, 0.5], respectively, for selected numbers of K intervals.

Table 4: rate overload with respect to the number of probability intervals for the uniform probability distribution of linear increment

Investigations in this section showed that discretizing the LPB probability interval (0, 0.5] over a small number of intervals with a fixed probability (eg, 8 to 10 intervals) has little impact on coding efficiency.

The entropy coding explained above for probability ranges thus allows individual coders to use fixed probabilities.

Next, we first show how a simple code for fixed probabilities can be designed. Given these results, algorithms are developed that jointly optimize the code design and the partition of the LPB probability interval (0, 0.5].

Entropic coding for fixed probabilities p = pik can be performed using arithmetic coding or variable length coding. For the latter case, the following approach appears to be simple and very efficient. It is considered a binary entropic coding scheme for which a variable number of binaries is mapped on a code word of variable length. For a unique decoding capability, the reverse mapping of a codeword to a sequence of binaries must be unique. And since you want to design a code that focuses the entropy limit as close as possible, we restrict the considerations to bijective mapping. Said bijective mapping can be represented by a binary tree where all the leaf nodes are associated with code words, as represented in Figure 15. The three edges represent binary events. In the example in Figure 15, the lower limits represent the LPB binary value and the upper limits represent the MPB binary value. The binary tree represents a prefix code for binaries if it is a complete binary tree, that is, if each node is either a leaf or has two descendants. Each leaf node is associated with a probability based on a given LPB p probability. The root node has the probability root = 1. The probability for all other nodes is obtained by multiplying the probability of the corresponding ancestor by p for the LPB descendants and q = 1 - p for the MPB descendants. Each leaf node Li is characterized by the number of edges LPB ai and the number of limits MPB bi from the root node to the leaf node. For a particular LPB probability p , the probability pi for the leaf node Li = {ai, bi} is equal to

pi = paI (1 - p) bI (B16) The binary tree T is totally characterized by the number of leaf nodes L and the associated pairs {ai, bi} where I = 0, ..., L-1.

Given the complete binary tree T and an LPB probability p, the optimal assignment of codewords to leaf nodes It can be obtained using the Huffman algorithm. The variable number resulting from mapping C bits to variable length codewords (V2V) is characterized by the number of codewords L, which is identical to the number of leaf nodes, and tuples {a /, b /, li } for I = 0, ..., L-1, where l / represents the length of the codeword that is associated with the corresponding leaf node L / = {a /, b /} . It should be noted that there are multiple possibilities for the codeword assignment given by the codeword lengths {l /} and the actual codeword assignment is not important as long as the codewords represent a uniquely decodable prefix code. The expected rate R (p, C) in bits per binary for a given C code and an LPB probability p is the ratio of the expected codeword length to the expected number of binaries per codeword

Code design is often limited by factors such as the maximum number of L codewords, the maximum number of binaries per codeword, or the maximum length of the codeword, or is restricted to particular structure codes (for example , to allow optimized analysis). If we assume that the set of codes usable for a particular application is given, the optimal code C * e Sc for a particular LPB probability p can be found by minimizing the expected rate R (p, C)

arg min R ( p, C )

C * (p) = vCeSc (B18)

As a faster alternative, minimization can also proceed on a given set of Sr binary trees and for each tree only one V2V C code is considered which is obtained by the Huffman algorithm. As an example, V2V codes are designed for various LPB p probabilities by considering all binary trees T for which the number of leaf nodes L is less than or equal to a given maximum Lm. In Figure 16, the relative rate increase p (p, C (p)) = R (p, C (p)) / N (p) is plotted on the LPB probability p for selected maximum table sizes Lm . The rate increase p (p) can normally be reduced by allowing larger table sizes. For higher LPB probabilities, a small table size L of 8 to 16 codewords is usually sufficient to keep the rate increment p (p) reasonably small, but for smaller LPB probabilities (for example, p <0.1) , larger table sizes L are required.

In the previous sections, optimal probability discretization assuming optimal codes and code design for fixed LPB probabilities was considered. But since, in general, the entropy limit cannot be achieved with actual V2V codes of limited table sizes, the code design and partitioning of the LPB probability interval (0, 0.5) must be considered together to obtain of an optimized entropic code design.

For a given interval / k = (p *, pk + 1], a code Ck of a given set Sc is a code c: optimal if you minimize the P 1 R ( p, C * ) f (p) dp

expected rate R = p * for the given interval.

_c : arg m _e in _{S c} P _{p *} +1 R ( p, C k) f ( _VC p) dp

(B19)

For practical designs, minimizing the integral in equation B19 can be simplified, with an impact

minimum in coding efficiency, first determining an optimal representative probability p'k for the

interval / * according to equation B10 and then choosing the optimal Ck code from the given set Sc

for the representative probability p'k according to equation B18.

Optimal interval edges p * can be deduced , with k = 1, K, K-1, given the set of codes C *, with k = 0, K, K-1, minimizing the expected global rate

K -1 í pk * 1

XJ R ( P, Ck ) f (p) dp

R = R ( {pk}, {Ck}) = k ^ p (B20)

8 _R

Set the first derivatives with respect to the edges of intervals equal to zero; SP ^k = 0, for k = 1, K, K-1, produces

Pk = pk with R ( Pk, Ck-1) = R ( Pk, Ck) (B21)

Similarly for equation B13, it can be shown that Pk is always an optimal solution, but depending on the probability distribution / (p) there may be additional optimal solutions. Therefore, an edge *

The optimal interval pk between two intervals Ik-1 and Ik with associated codes Ck-1 and Ck given, respectively, is the point of intersection of the functions R (p, Ck-1) and R (p, Ck).

Consequently, the following interactive algorithm can be used to jointly deduce the probability interval partition and associated codes given the number K of probability intervals, the set of possible Sc codes, and the probability distribution / (p), with pe ( 0, 0.5].

Algorithm 2:

1) Deduce the initial bounds of the probability interval pk, with k = 0, K, K, using algorithm 1 specified in section 3

2) Derive the representations pik for the probability intervals Ik, with k = 0, K, K-1, according to equation B10

3) Derive the codes Ck and Sc for the representations of interval pik, with k = 0, K, K-1, according to equation B18

4) Update the interval edges pk, with k = 1, K, K-1, according to equation B21

5) Repeat the three previous stages until convergence

Steps 2 and 3 in algorithm 2 could also be replaced by a direct deduction of the codes Ck and Sc, with k = 0, K, K-1, based on the edges of the interval pk, with k = 0, K, K , according to equation B19. And, as mentioned in section 4.1, minimization in step 3 can also proceed on a given set of binary trees S ^t where for each binary tree T only one V2V Ck code obtained by the Huffman algorithm is considered.

As an example, the partition at K = 12 probability intervals and corresponding V2V codes are derived together using algorithm 2. To do this, minimization in step 3 of the algorithm was replaced by equivalent minimization on a given set of binary trees Sr where the evaluated code C for each T tree was obtained using the Huffman algorithm. We consider T trees with a maximum number of Lm = 65 leaf nodes and therefore C codes with up to 65 table entries. All T binary trees with up to 16 leaf nodes have been evaluated in minimization; For trees with more than 16 leaf nodes, a suboptimal search was used given the best results for trees with a smaller number of leaf nodes.

In Figure 17, the expected rate increase relative to the entropy limit AR (p) = R (p) - H ( p) for the sample code design is plotted on the LPB p probability. For comparison, the expected AR rate increase for the theoretically optimal probability interval discretization (as discussed in section 3) and the theoretically optimal probability discretization with the additional constraint pik-1 = 0 are also plotted within the diagram. ,5. It can be seen that the discretization of the joint probability interval and the design of the V2V code leads to a displacement of the edges of the interval (the edges of the interval pk, with k = 1, K, K-1, are given by the local maximum of the curves AR (p)). The relative expected global rate increase with

ratio to the entropy limit for the design example with real V2V codes is p = 0.24%, when assuming a uniform probability distribution / (p). The corresponding relative rate increments for theoretically optimal probability interval discretization and theoretically probability discretization

optimal with the additional constraint pik-1 = 0.5 are p = 0.12% and p = 0.13%, respectively.

The finalization of the code word can be done as follows. When encoding a finite sequence of symbols {sq, K, sw-1}, each of the K. binary encoders processes a finite sequence of binaries of DC

encoding bk = { 0, K, b ^ _x } k, with k = 0, K, K-1. And it has been ensured that; for each of the K encoders bc bc

binaries, all encoding binaries of sequence k = {0, K, b ^ _i} k can be reconstructed given the code word or code word sequence Ck ( ^b k ^c ).

When arithmetic encoding is used, the arithmetic codeword for the encoding binary sequence has to be finalized in a way, in which all the encoding binaries can be decoded given the codeword. For the V2V codes described above, the binaries at the end of the sequence ^b k ^c may not represent a sequence of binaries that is associated with a code word. In such a case, any code word that contains the rest of the binary sequence can be written as a prefix. The overhead can be minimized by choosing the corresponding code word (or one of those code words) that has the minimum length. On the decoder side, binaries read in addition to the end of the binary sequence are discarded, which can be identified given the bitstream syntax and binary conversion schemes.

Below is a simple example code layout. For illustration purposes, consider the simple example of a {s} font with three letters and fixed associated probabilities of ps ( aa) = 0.7, ps (ai) = 0.18, and Ps (a2) = 0, 12. The corresponding ternary choice tree can be converted to a complete binary tree as shown in Fig. 18.

A conversion to binary for the full binary tree in Fig. 18 is given in Table 5. The ternary symbol pmf ps becomes two binary pmfs pb ⁰ = (0.7, 0.3) and pb ¹ = (0 , 6, 0.4). For each symbol s in the bitstream, the bü binary is present. When bü equals 0, b1 is also present. Note that the conversion to binary given in Table 2 is identical to an optimal single-letter Huffman code for font s.

Table 5: Conversion to binary of a three letter font. The LPB plpb probabilities are 0.3 for the first in ri 4 rln in ri

The entropy for source s is

H (0.7, 0.18, 0.12) = H (0.7, 0.3) 0.3 H (0.6, 0.4) = 1.1726 bit / symbol (B22)

The average length of the single letter Huffman code word is given as

M -1

1l p ^ Hc

lHC =, = ° = 1, 3 bits / symbol (B23) corresponding to a redundancy of ^phc = 0.1274 bit / symbol or 10.87% expected rate overhead.

For the example of conversion to particular binary with fixed pmfs, binaries b0 and b1 already represent binary encoding, since for both binaries the LPB bf PB value is equal to 0. The distribution f (s) of the LPB probabilities is discrete , with / (p) = 0 except for p = 0.3 and p = 0.4. Consequently, the optimal probability discretization leads to K = 2 intervals with the representations p «= 0.3 and pn = 0.4. The edge of the interval p1 between these intervals can be arbitrarily chosen in [0.3, 0.4).

For source encoding, the source symbol sequence is converted to binary into a binary sequence. The b0 binary is transmitted for each source symbol. The binary b1 is only transmitted when b0 = 0. The binary b0 and b1 are encoded separately with constant LPB probabilities of p «= 0.3 and bn = 0.4, respectively.

Efficient coding of a fixed probability binary alphabet can be achieved simply by mapping a V2V. Examples of mapping for V2V with small coding tables for the LPB probabilities ^plpb = 0.3 and ^plpb = 0.4 are given in Table 6 and Table 7, respectively. V2V mapping for ^plpb = 0.3 produces a redundancy of 0.0069 bit / binary or 0.788%. For the LPB probability of ^plpb = 0.4, the redundancy is 0.0053 bit / binary or 0.548%.

Table 6: Tree and binary codes for an LPB probability of ^plpb = 0.3. The redundancy of this code is 0,788

%

Table 7: Tree and binary codes for an LPB probability of Plpb = 0.4. The redundancy of this code is 0.548

%

The overall expected rate incurred by the new coding method is

£ nc = £ bi + 03 ■ £ b

= 1,181 bit / symbol (B24)

The global redundancy is 0.73% relative to the entropy limit, which represents a significant improvement compared to the Huffman single letter code.

It could be argued that a similar encoding efficiency improvement could be obtained by creating a serial length code. For the example above, we can construct a series length code for the most likely symbol by considering the series of up to two symbols. Each of the events {a0a0, aüai, aüa2, ai, a2} would be associated with a separate codeword. This code produces the redundancy of 1.34% in relation to the entropy limit. Actually, V2V codes can be considered as a generalization of the serial length codes for binary symbols (the V2V code in Table 3 effectively represents a serial length code). For a single symbol alphabet with fixed probabilities, coding efficiency similar to that of the presented approach can be achieved by creating a code that maps a variable number of source symbols to code words of variable length. The main advantage of the presented approach is its flexibility in mapping arbitrary source symbol sequences with fixed or adaptive probability estimates to a small number of simple binary coders that operate with fixed LPB probabilities. How to achieve unique decoding capability is considered below.

With the entropy encoding scheme presented, the encoding of a source symbol sequence s = {s0, K, sw-1} consists of the following three basic steps.

conversion to binary of the symbol b = {bü, K, bs-1} = Yb ( s) that drive the binary sequence b = {b0, K, bB-1} conversion of the binary sequence into a binary sequence of encoding bc = { ^b 0 ^c , K, ^b 8 ^c -1} = Yc (b) bc binary entropic encoding of the sequence of encoding binaries b ^c = { ^b 0 ^c K, B-1} using discretization of the range of probability and K fixed binary encoders

The symbol sequence s = {Sü, K, sw-1} can be decoded only, if the sequence of encoding binaries bc = { ^b 0 ^c , K, ^bc 8-1} can be decoded only and the Yb and Yc maps can reverse.

Ye reporting the encoder mapping of a sequence of one or more encoding binaries b ^c = { ^bc 0, K} over a sequence of one or more codewords c ( b c) = {c0, ...}

c (bc) = Ye ( b c) (B25) For a unique decoding capability of a bc- encoding binary sequence given the sequence of c ( bc) codewords , the Ye encoder mapping should have the property that a single c ( b c) codeword is assigned to every possible sequence of bc encoding binaries :

This property is satisfied whenever arithmetic codes or prefix codes are used. It is particularly suited for V2V codes described in section 4.1 (including the completion of the codeword described in section 4.3), since V2V codes represent prefix codes for variable numbers of binaries.

However, in the entropic coding approach presented, the bc coding binary sequence is partitioned into K subsequences ^b k ^c , with k = 0, ..., K-1,

b ^{0 c bc c}

{K, bK-1} = Yp (bc) (B27)

and each of the subsequences ^b k ^c is assigned a sequence of code words ck ( ^b k ^c ) using a particular yk encoder mapping. Accordingly, the single decoding capability condition has to be extended. A sequence of bc encoding binaries is decodable only given K codeword sequences ck ( ^b k ^c ), with k = 0, K, K-1, if each subsequence of ^bc k encoding binaries is uniquely decodable given the word of corresponding ck ( ^b k ^c ) code and partition rule Ye is known to the decoder. The partition rule Ye is given by the discretization of the probability interval LPB {Ik} and the bc

LPB HLPB probabilities that are associated with encoding binaries 1 , with j = 0, K, B-1. Therefore, the discretization of the probability interval LPB {Ik} has to be known on the decoder side and the probability bc

LPB Plpb for each bj encoding binary, with j = 0, K, p -1, has to be deduced in the same way on the encoder and decoder side.

For Yc mapping of a binary sequence into a c-encoding binary sequence, each unique bj, with j = 0, K, B-1, is converted by binary mapping j = c jj LPB . On the decoder side, the sequence of binaries can be deduced by binary mapping

bj = ( ^yc )>; ) = bc © bjp (B28)

with j = 0, K, B-1. If the LPB value ^bl for each binary bj is deduced in the same way on the encoder side as that of the decoder, these mappings ( ^yc r 1 represent the inverses of the corresponding ^yc encoder mappings and therefore

b ° ¡ © bLPB - bj © bLPB © bLPB _ bj © 0 = b (B29)

and therefore, the conversion Yb of a sequence of binaries b into a sequence of encoding binaries bc is invertible.

Finally, the inversion capacity of the conversion to binary b = Yb (s) is investigated by means of which each symbol s, with i = 0, K, N-1, is mapped on a binary sequence tí = Yb ( s' ). A symbol s¡ can be decoded only given the corresponding binary sequence t í if the binary conversion mapping Yb assigns a b j ^q *

binary sequence m different from each letter m of alphabet Ai for symbol s. However, this condition is not sufficient, since the partition of the binary sequence b = {bo, K, bB-1} into binary sequences t í that correspond to the symbols Si, with i = 0, K, N -1, is not known to the decoder. A sufficient condition is b j q *

gives when for each symbol s, the binary sequences m that are associated with the letters m of the corresponding alphabet Ai form a prefix code and the binary conversion maps y'h for each symbol s, with i = 0, K, N-1, are known on the decoder side.

The unique decoding capability conditions for the presented entropic encoding approach can be summarized as follows:

• Yb binary conversion mappings represent prefix codes and are known to the decoder (in symbol encoding order)

b j pj

• the probability models ( LPB , rLPB ) for all bj binaries are deduced in the same way on the encoder side as that of the decoder

• the partition of the probability interval LPB (0, 0.5] in K intervals Ik, with k = 0, K, K-1, is known to the decoder

• the mapping ^Y 'c ^k for each probability interval Ik, with k = 0, K, K-1, represents a uniquely decodable code

Examples of the overall design of the encoder and decoder are described in more detail below. We will focus on encoding schemes, in which the {bLPB, P ^lpb } probability models for binaries are estimated directly on the encoder and decoder side and the K binary encoders use V2V mappings described above. Each source symbol s shall be associated with a symbol category cs, which determines the type of symbol including its range of values. The order of symbols and associated symbol categories should be given by the syntax, which is assumed to be known on the encoder and decoder side.

The block diagram for an example PIPE encoder and PIPE decoder design is illustrated in Figure 19. On the encoder side, symbols s with associated symbol categories Cs are fed into the

converter to binary, which converts each symbol s into a sequence of binaries bs = ^Y 'b ^{° s} (s).

The binary conversion scheme ^Y 'b ^Cs is determined based on the symbol category Cs. In addition, the binary converter associates each binary b with a sequence of binaries, with an indication Cb of the probability model, which specifies the probability model used for encoding binary b. The indication Cb of the probability model can be derived based on the symbol category Cs, the binary number of the current binary within the sequence binary s, and / or the values of binaries and symbols already encoded.

The probability estimator and allocator maintains multiple probability models, which are characterized by pairs of values {bLPB, p ^lpb }. It receives binaries b and indications Cb from the probability model of the converter to binary, and sends the value LPB bLPB and the probability LPB p ^lpb of the indicated probability model to the part that deduces the encoding binary and the probability quantizer, respectively. Subsequently, the corresponding probability model {bLPB, p ^lpb } is ^updated using the received binary b value.

The part that deduces the binary receives binaries b and associated LPB bLPB values from the binary converter and the probability estimator and allocator, respectively, and sends bc encoding binaries , which are deduced by bc = b® bLPB, to the probability quantizer . The probability quantizer sends each bc encoding binary to one of the K binary encoders. Contains information about the quantization {Ik} of the LPB probability interval. The LPB probability p ^lpb , which is associated with a bc encoding binary and received from the probability estimator and allocator, is compared to the edges of the interval {pk} and the index of the probability interval k, for which it is deduced ^plpb and Ik. The bc encoding binary is then sent to the associated binary encoder.

Each of the K binary encoders consists of a binary buffer and a binary encoder. Each binary buffer receives bc encoding binaries from the probability quantizer and stores them in the encoding order. The binary encoder implements a particular V2V mapping and compares the sequence of binaries in the binary buffer with the binary sequences that are associated with codewords. If the binary sequence in the binary buffer equals one of those binary sequences, the binary encoder removes the binary sequence {bc} from the binary buffer and writes the associated codeword ({bc} ) in the corresponding codeword stream. At the end of the encoding process for a symbol sequence, for all binary encoders for which the binary buffers are empty, a completion codeword is written as described in section 4.3.

The resulting K codeword streams can be transmitted separately, packaged or stored, or can be interleaved (compare section 6.2) for the purpose of transmission or storage.

On the decoder side, each of the K binary decoders consisting of a binary decoder and a binary buffer receives a stream of code words. The binary decoder reads the codewords ({bc}) from the codeword stream and inserts the associated binary sequence {bc} in the encoding order, into the binary buffer.

Decoding the symbol sequence is triggered by the underlying syntax. Requests for an s symbol along with the category of the cs symbol are sent to the binary converter. The converter to binary converts these token requests into binary requests. A request for a binary is associated with an indication Cb of the probability model, which is derived in the same way as in the encoder, and is sent to the probability estimator and allocator. The probability estimator and allocator works similarly to its counterpart on the encoder side. Based on the Cb indication of the probability model, it identifies a probability model and sends its LPB bLPB value and LPB probability p ^lpb to the part that deduces the binary and to the probability quantizer, respectively.

The probability quantizer determines one of the K binary decoders based on the probability LPB P ^lpb , in the same way that the binary encoder determines it on the encoder side, it removes the first encoding binary bc in the encoding order, from the corresponding binary buffer, and sends it to the part that deduces the binary. The part that deduces the binary receives the bc encoding binaries and associated LPB bLPB values from the probability quantizer and probability estimator and allocator, respectively, and determines the binary values b = bc® bLPB . As a final response to the binary request sent by the binary converter, the binary-deducing party sends the decoded binary value b to the binary converter and the probability estimator and allocator.

In the probability estimator and allocator, the value of the decoded binary b is used to update the probability model {bLPB, P ^lpb }, which was chosen using the associated cb value, in the same way as on the encoder side. Finally, the binary converter adds the received binary b to the sequence of binaries s that has already been received for a symbol request and compares this sequence of binaries s with the binary sequences that are

associated with symbol values ^yCs or by the binary conversion scheme 1 b . If the binary sequence s matches one of those binary sequences, the corresponding decoded symbol s is output as a final response to the symbol request. Otherwise, the binary converter sends additional binary requests until the symbol s is decoded.

Decoding of a symbol sequence is terminated if no further requests for additional symbols are received, which are triggered by the syntax. Any bc- encoding binaries that may be contained in the binary buffers at the end of the entropic decoding process (as a result of the completion codewords) are discarded.

After having described certain embodiments for the PIPE encoders and PIPE decoders in Figs. 1 and 2 with respect to Figs. 3 to 13 and having provided a mathematical basis in relation to PIPE coding in general in relation to Figs. 14 to 19, embodiments are described in more detail for entropic encoding and decoding apparatus, relative to Figs. 20-24. The following embodiments of Figs. 22 to 24 interleave with each other not only PIPE encoded bitstreams, but they interleave the VLC bitstream and the PIPE encoded bitstreams together. Compared thereto, the embodiments for PIPE encoders and PIPE decoders with PIPE encoded bitstream interleaving, in particular the embodiments of Figs. 7 to 13 simply provide separate interleaving of the PIPE encoded bitstreams. As already mentioned above, even these embodiments can serve as the basis for achieving a fully interleaved bitstream by using another interleaver / deinterleaver pair 134 and 228, respectively (see Figs. 1 and 2), such as, for example, by using the bitstream interleaver as shown in Figs. 5 and 6. However, the embodiments described below perform interleaving at once in both the VLC bitstream and the PIPE encoded bitstreams with the use, in other words, of an interleaver / deinterleaver in one step 128 and 230, respectively.

Before describing in detail the embodiments where VLC and PIPE encoded symbols are interleaved within a bitstream, in order to achieve a more adequate compromise between complexity and encoding efficiency, a basic structure thereof is described without interleaving with respect to Figs. 20 and 21.

The structure of the entropy coding apparatus of Fig. 1a is shown in Fig. 20. The entropic coding apparatus converts a stream of source symbols 1 to, corresponding to the combination of VLC encoded and PIPE encoded source symbols of Figs. . 1 and 2, specifically 106 and 218, respectively, in a set of two or more partial bit streams 12, 12 a, bit stream 12 corresponding to bit streams 112 and 206 of Figs. 1 and 2.

As already indicated, each source symbol 1a may have associated with it an indication that specifies whether the source symbol is encoding using standard VLC codes within VLC encoder 22a, which corresponds to VLC encoder 102 in Fig. 1, or whether the source symbol is to be encoded with a PIPE encoding concept. As already described above with respect to Figs. 1 and 2, this indication cannot be explicitly transmitted to the decoding side. Rather, the associated indication may come from the type or category of the same source symbol.

VLC encoded symbols 1b are encoded with standard VLC codes, which in turn may depend on the recently mentioned symbol category or symbol type using a VLC encoder 22a. The corresponding code words 11a are written in a different partial bitstream 12a. Non-VLC code symbols 1 are encoded using PIPE encoding as described above with respect to Figs. 1 and 3, for example, whereby multiple partial bit streams 12 are obtained. Some of the source symbols 1a may have already been converted to binary in a manner consisting of the two-part binary conversion already mentioned above with respect to to Fig. 1a. One of these parts can be encoded with the PIPE approach and written to the corresponding partial bitstreams 12. The other part of the binary sequence can be encoded with the standard VLC codes and written to the corresponding partial bitstream 12a.

The basic entropic decoding apparatus conforming to the embodiment of Fig. 20 is shown in Fig. 21. The decoder basically performs the reverse operations of the encoder of Fig. 20, so that the previously encoded sequence of source symbols 27, 27a is decoded from a set of two or more partial bit streams (24, 24a). The decoder includes two different process streams: a stream for data requests, which replicates the encoder's data stream, and a data stream, which represents the inverse of the encoder's data stream. In the illustration in Fig. 21, the dashed arrows represent the flow of data requests, while the solid arrows represent the flow of data. The decoder building blocks basically replicate the encoder building blocks, but implement the inverse operations.

In a preferred embodiment of the invention, each symbol request 13a is associated with an indication that specifies whether the source symbol is encoded using standard VLC codes or with the concept of PIPE encoding. As already mentioned above with respect to Fig. 20, this indication may come from parsing rules or syntax of syntax elements represented by the same source symbol. For example, with respect to Figs. 1 and 2, it has been described that different types of syntax elements can be associated with different encoding schemes, namely VLC encoding or PIPE encoding. The same can be applied for different parts of binary conversions, or more generally, other symbolizations of syntax elements. If a symbol is VLC encoded, the request is passed to VLC decoder 22a and a VLC code word 23a is read from a different partial bitstream 24a. The output of the corresponding decoding symbol 27a occurs. If a symbol is encoded with PIPE, the symbol 27 is decoded from a set of partial bitstreams 24 as described above with respect to Fig. 4, for example. In another preferred embodiment of the invention, some of the source symbols are converted to binary in such a way that the conversion to binary consists of two parts. One of these parts is encoded with the PIPE approach and correspondingly decoded from the associated partial bitstreams 24. And the other part of the binary sequence is encoded with standard VLC codes and decoded with a VLC decoder 22a that reads the corresponding code words 23a from a different partial bitstream 24a.

Transmission and multiplexing of partial bit streams (VLC encoded and PIPE encoded)

The partial bit streams 12, 12a that are created by the encoder can be transmitted separately, or they can be multiplexed into a single bit stream, or the code words of the partial bit streams can be interleaved into a single bit stream.

In a preferred embodiment of the invention, each partial bitstream for a quantity of data is written to a data packet. The amount of data may be an arbitrary set of source symbols such as a still image, a field or a frame from a video sequence, a fragment from a still image, a fragment from a field or a frame from a video sequence, or a frame of audio samples, etc.

In another preferred embodiment of the invention, two or more of the partial bit streams 12, 12a for a data amount or all the partial bit streams for a data amount are multiplexed into a data packet. The structure of a data packet containing multiplexed partial bit streams can be as illustrated in Fig. 5.

Data packet 300 consists of a header and a partition for the data of each partial bitstream (for the amount of data considered). The data packet header 301 contains indications for partitioning the (remainder) of the data packet into data segments 302 of the bit stream. Along with the directions for partitioning, the header may contain additional information. In a preferred embodiment of the invention, the indications for partitioning the data packet are the locations of the beginning of the data segments in units of bits or bytes or multiples of bits or multiples of bytes. In a preferred embodiment of the invention, the locations of the beginning of the data segments are encoded as absolute values in the header of the data packet, either relative to the start of the data packet or relative to the end of the header or relative to at the beginning of the previous data package. In a preferred embodiment of the invention, the locations of the beginning of the data segments are differentially encoded, that is, only the difference between the actual beginning of a data segment and a prediction for the beginning of the data segment is encoded. The prediction can be deduced based on already known or transmitted information such as the overall size of the data packet, the size of the header, the number of data segments in the data packet, the location of the beginning of the preceding data segments. In a preferred embodiment of the invention, the location of the beginning of the first data packet is not encoded, but is deduced based on the size of the header of the data packet. On the decoder side, the transmitted partition indications are used to deduce the beginning of the data segments. The data segments are then used as partial bitstreams 12, 12a, and the data contained in the data segments is fed into the corresponding binary decoders and VLC decoders in sequential order.

Codeword interleaving (VLC and PIPE codewords)

For some applications, the above-described multiplexing of partial bit streams (for a number of source symbols) in a data packet may have the following disadvantages: on the one hand, for small data packets, the number of bits for the information Secondary that is required for partitioning signaling can become significant relative to the actual data in the partial bitstreams, ultimately reducing the encoding efficiency. On the other hand, multiplexing may not be suitable for applications that require low delay (for example, for video conferencing applications). With the described multiplexing, the encoder cannot initiate the transmission of a data packet before the partial bit streams have been fully created, since the locations of the partition start are not known before. Additionally, in general, the decoder has to wait until it receives the start of the last data segment before it can start decoding a data packet. For applications such as video conferencing systems, these delays can be added to an additional overall system delay of multiple video images (in particular, for bit rates approaching the transmission bit rate and for encoders / decoders that require virtually the time interval between two images for encoding / decoding an image), which is critical for such applications. To overcome the disadvantages for certain applications, the encoder of a preferred embodiment of the invention can be configured in such a way that the code words that are generated by the two or more binary encoders and the VLC encoder are interleaved in a single stream of bits. The bitstream with the interleaved codewords can be sent directly to the decoder (when a small buffer delay is neglected, see below). On the decoder side, the two or more binary decoders and the VLC decoder read the codewords directly from the bitstream in the decoding order; decoding can be started with the first bit received. Furthermore, no secondary information is required for signaling the multiplexing (or interleaving) of the partial bit streams.

The basic structure of an encoder with codeword interleaving is shown in Fig. 22. The binary encoders 10 and the VLC encoder 10a do not write the codewords directly into the partial bitstreams, but connect to a single codeword buffer 29, from which codewords are written to bitstream 34 in encoding order. Binary encoders 10 send requests for one or more codeword 28 entries to codeword buffer 29 and then send codewords 30 to buffer 29, which are stored in reserved buffer entries . The VLC encoder 10a directly writes the VLC codewords 30a into the codeword buffer 29. Codewords 31 (generally of variable length) from the codeword buffer 29 are accessed by a word writer. code 32, which writes the corresponding bits 33 into the produced bit stream 34. Codeword buffer 29 functions as a first-in-first-out memory; codeword entries that are reserved before are written to the bitstream earlier.

In a preferred embodiment of the invention, the codeword buffer functions as follows. If a new binary 7 is sent to a particular binary 8 buffer and the number of binaries already stored in the binary buffer is zero and no codeword is currently reserved in the codeword buffer for the encoder of binary that is connected to the particular binary buffer, the connected binary encoder 10 sends a request to the codeword buffer, whereby one or more codeword entries are reserved in the word buffer code 29 for the particular 10 binary encoder. Codeword entries can have a variable number of bits; an upper threshold for the number of bits in a buffer entry is normally given by the maximum codeword size for the encoder. corresponding binary. The next codeword or the following codewords that are produced by the binary encoder (for which codeword input or codeword entries have been reserved) are stored in the reserved input or entries of the codeword buffer. If all reserved buffer entries in the codeword buffer for a particular binary encoder are filled with codewords and the next binary is sent to the binary buffer that is connected to the particular binary encoder, one or more new codewords are reserved in the codeword buffer for the particular binary encoder, etc. The VLC encoder 10a writes the VLC codewords 30a directly to the next free input of the codeword buffer 29, i.e. for the VLC encoder the codeword reservation and the writing of the codewords is done once. Codeword buffer 29 represents a first-in-first-out buffer in a certain way. Entries to the buffer are reserved in sequential order. Codewords for which the corresponding buffer entries have been reserved before are written to the bitstream earlier. Codeword writer 32 checks the status of codeword buffer 29 both continuously and after codeword 30 is written to codeword buffer 29. If the first entry to memory buffer contains a complete codeword (i.e. the buffer entry is not reserved, but includes a codeword), the corresponding codeword 31 and the corresponding buffer entry are removed from the buffer of codewords 20 and codeword 33 bits are written to the bitstream. This process is repeated until the first buffer entry does not contain a codeword (that is, it is reserved or free). At the end of the decoding process, that is, if all source symbols of the considered amount of data have been processed, the codeword buffer should be purged. For that purge process, the following apply to each binary / binary encoder buffer as a first step: if the binary buffer does not contain binaries, a binary with a particular value or arbitrary value is added until the resulting binary sequence represents a sequence of binaries that is associated with a codeword (as noted above, a preferred way to add binaries is to add those binary values that produce the shortest possible codeword - or one of them - which is associated with a sequence of binaries containing the original content of the binary buffer as a prefix), then the codeword is written to the next reserved buffer entry for the corresponding binary encoder and flushed ( and the corresponding) binary buffer. If more than one buffer entry has been reserved for one or more binary encoders, the codeword buffer may still contain reserved codeword entries. In that case, these codeword entries are populated with arbitrary, but valid, codewords for the corresponding binary encoders. In a preferred embodiment of the invention, the shorter valid codeword or one of the shorter valid codewords (if there are multiple) is inserted. The VLC encoder does not require any completion. Finally, all the remaining codewords in the codeword buffer are written to the bitstream.

Two examples for the status of the codeword buffer are illustrated in Fig. 23. In example (a), the codeword buffer contains 4 entries that are populated with a codeword (two of which are VLC entries) and 3 reserved entries. In addition, the next entry in the free buffer is marked. The first entry is populated with a code word (that is, binary encoder 2 just writes a code word to a previously reserved entry). In the next stage, this codeword will be removed from the codeword buffer and written to the bitstream. Then the first codeword reserved for binary encoder 3 is the first buffer entry, but this entry cannot be removed from the codeword buffer, as it is only reserved, but has not been written. code word in this entry. In example (b), the codeword buffer contains 4 entries that are populated with a codeword (one of which is a VLC buffer entry) and 4 reserved entries. The first entry is marked as reserved and therefore the codeword writer cannot write a codeword to the bit stream. Although 4 codewords are contained in the codeword buffer, the codeword writer has to wait until the codeword is written to the first buffer entry reserved for binary encoder 3. Note that codewords must be written in the order in which they were reserved, to be able to reverse the process on the decoder side (see below). And additionally note that a VLC buffer entry is always full, since the reservation and writing of the codeword is done at once.

The basic structure of a codeword interleaving decoder is shown in Fig. 24. The binary decoders 22 and the VLC decoder 2a do not read the codewords directly from separate partial bit streams, but instead connect to a memory bit buffer 38, from which code words 37, 37a are read in the coding order. It should be noted that bit buffer 38 is not necessarily required, since codewords could also be read directly from the bit stream. Bit buffer 38 is mainly included in the illustration to clearly separate different aspects of the processing chain. Bits 39 of bit stream 40 with embedded codewords are They sequentially insert into bit buffer 38, which represents a first-in-first-out buffer. If a particular binary decoder 22 receives a request for one or more binary streams 35, binary decoder 22 reads one or more codewords 37 from bit buffer 38 through bit request 36. The decoder You can instantly decode the source symbols. Similarly, if VLC decoder 22a receives a request for a new symbol 19a it reads the corresponding VLC code word 37a from bit buffer 38 and returns the decoded symbol 27a. Note that the encoder (as described above) must ensure by properly operating the codeword buffer that the codewords are written in the same order to the bitstream in which they are requested by the decoders. binary. In the decoder, the entire decoding process is triggered by requests for source symbols. Parameters such as the number of codewords that are reserved on the encoder side by a particular binary encoder and the number of codewords that are read by the corresponding binary decoder must be the same.

Variable length codeword interleaving with low delay constraints

The described codeword collation does not require that any partition information be submitted as secondary information. And since the codewords are embedded in the bitstream, the delay is generally small. However, there is no guarantee that a particular delay constraint will be obeyed (for example specified by a maximum number of bits that are stored in the codeword buffer). Additionally, the buffer size required for codeword buffering can theoretically be made very large. When considering the example in Fig. 23 (b), it would be possible that no additional binaries are sent to the binary 3 buffer and therefore the binary 3 encoder will not send any new codeword to the word buffer code until the purge process is applied to the end of the data packet. So all the code words for binary encoders 1 and 2 would have to wait until the end of the data packet, before they can be written to the bitstream. This drawback can be overcome by adding an additional mechanism to the encoding process (and also to the decoding process as described below). The basic concept of this additional mechanism is that if a measure in relation to the delay or an upper limit of the delay (see below) exceeds a specified threshold, the first entry in the reserved buffer is filled by purging the buffer from corresponding binary (using a mechanism similar to the one at the end of the data packet). Using such a mechanism, the number of standby buffer entries is reduced until the associated delay measure is less than the specified threshold. On the decoder side, binaries that have been inserted on the encoder side to obey the delay constraint must be discarded. Basically the same mechanism can be used for this discard of the binaries as on the encoder side.

After having described in detail possibilities for interleaving of the VLC and PIPE encoding bitstreams, the description now again focuses on the previously mentioned syntax elements decomposed into source symbols as mentioned with respect to the Figs. 1b, 1c and 2b. For purposes of illustration, the following description assumes that the syntax elements thus decomposed are absolute transform coefficient levels. However, this is only an example, and other types of syntax elements can be handled in the same way. In particular, the encoding of absolute levels by partitioning and using different entropy codes in block-based image and video encoders is described below.

For example, images in the video stream are typically broken down into blocks. Blocks or block color components are predicted by either compensated motion prediction or intra prediction. The blocks can have different sizes and can be either quadratic or rectangular. All samples of a block or a block color component are predicted using the same set of prediction parameters, such as reference indices (which identify a reference image in the set of already encoded images), motion parameters (which specify a measure for the movement of the blocks between a reference image and the current image), parameters for the specification of the interpolation filter, intra prediction modes, etc. Motion parameters can be represented by displacement vectors with a horizontal and vertical component or by higher order motion parameters such as cognate motion parameters consisting of 6 components. It is also possible that more than one set of prediction parameters (such as benchmarks and motion parameters) are associated with a single block. In that case, for each set of prediction parameters, a single intermediate prediction signal is generated for the block or the color component of a block, and the final prediction signal is constructed using a weighted sum of the intermediate prediction signals . The weighting parameters and potentially also a constant offset (which is added to the weighted sum) can be either fixed for an image, or a reference image, or a set of reference images, or can be included in the parameter set prediction for the corresponding block. Similarly, still images are also frequently broken down into blocks, and blocks are predicted by an intra-prediction method (which can be either a spatial intra-prediction method or a simple intra-prediction method that predicts the continuous block component) . In the case of a corner, the prediction signal can also be zero.

The difference between the original blocks or the color components of the original blocks and the corresponding prediction signals, which are also called residual signals, are normally transformed and quantized. A two-dimensional transform is applied to the residual signal and the resulting transform coefficients are quantized. For this transform encoding, the blocks or the color components of the blocks, for which a particular set of prediction parameters has been used, may be further partitioned prior to the application of the transform. The transform blocks can be equal to or smaller than the blocks that are used for prediction. It is also possible that a transform block includes more than one of the blocks that are used for prediction. Different transform blocks in a still image or a frame in a video sequence can have different sizes, and the transform blocks can represent quadratic or rectangular blocks.

All of these prediction and residual parameters can form the flow of syntax elements 138 and 226, respectively.

The resulting quantized transform coefficients, also referred to as transform coefficient levels, can then be transmitted using entropic encoding by one of the above encoding schemes. To this end, a block of transform coefficient levels can be mapped onto a vector (i.e., an ordered set) of transform coefficient values using one scan, where different scans can be used for different blocks. A zigzag scan is often used. For blocks containing only samples of a field from an interlaced frame (these blocks can be blocks in coded fields or blocks of fields in coded frames), it is also common to use a different scan designed specifically for field blocks. A possible encoding scheme for encoding the ordered sequence resulting from transform coefficients is encoding at the serial level. Typically, a large number of transform coefficient levels are zero, and a set of successive transform coefficient levels that are equal to zero can be efficiently represented by encoding the number of successive transform coefficient levels that are equal to zero ( the string) by a respective syntax element. For the remaining transform coefficients (non-zero), the actual level is encoded in the form of respective syntax elements. There are several code alternatives at the serial level. The string before a non-zero coefficient and the level of the non-zero transform coefficient can be coded together using a single syntax element. Frequently, special syntax elements are included for the end of the block, which is sent after the last non-zero transform coefficient. Or it is possible to code the number of non-zero transform coefficient levels first, and depending on this number, the levels and series are coded.

A somewhat different approach is used for highly efficient CABAC entropic encoding in H.264 / AVC. In this case, the coding of the transform coefficient levels is divided into three stages. In the first stage, a binary coded_key_indicator binary syntax element is transmitted for each transform block, signaling whether the transform block contains significant transform coefficient levels (i.e. transform coefficients that are non-zero). If this syntax element indicates that significant transform coefficient elements are present, a significance map is encoded with a binary value, which specifies which of the transform coefficient levels has non-zero values. And then, in an inverse scanning order, the values of the non-zero transform coefficient levels are encoded. The significance mapping is encoded in a syntax element stream 138 as follows. For each coefficient in the scanning order, a significant binary_coefficient_coding syntax element is encoded, which specifies if the corresponding transform coefficient level is not equal to zero. If the binary of the significant_coefficient_indicator is equal to one, that is, if a non-zero transform coefficient level exists at this scan position, an additional binary syntax element last_significant_coefficient_indicator is encoded. This binary indicates whether the significant transform coefficient level is the last significant transform coefficient level within the block or whether additional significant transform coefficient levels are still in the scanning order. If last_significant_coefficient_indicator indicates that there is no additional significant transform coefficient to follow, no additional syntax elements are encoded for the significance map specification for the block. In the next stage, the values of the significant transform coefficient levels, whose locations within the block have already been determined by the significance map, are coded. Values of the significant transform coefficient levels are encoded in reverse scan order using the following three syntax elements. The binary syntax element coefficient_absolute_major_one indicates whether the absolute value of the level of significant transform coefficient is greater than one. If the binary syntax element coefficient_absolute_higher_one indicates that the absolute value is greater than one, an additional syntax element coefficient_absolute_level_minus_two is sent specifying the absolute value of the transform coefficient level to minus two. This is the class of syntax elements whose processing is performed according to Fig. 1b, 1c and 2b in the following embodiment. Finally, the binary syntax element indicator_sign_coefficient, which specifies the sign of the transform coefficient value, is encoded for each significant transform coefficient level. It should be noted again that the syntax elements referring to the significance map are encoded in scan order, whereas the Syntax elements that refer to the current values of the transform coefficient levels are encoded in reverse scan order allowing the use of more appropriate context models. An adaptive scan pattern may also be used for the significance map as in the first H.265 / HEVC test model. Another concept is used for coding the absolute transform coefficient levels for the larger 4x4 block transform in the first H.265 / HEVC test model. In case of transform blocks larger than 4x4, the larger transform block is partitioned into 4x4 blocks and the 4x4 blocks are encoded in the scan order while the reverse scan order is used for each of the 4x4 blocks.

In CABAC entropic encoding in H.264 / AVC, all syntax elements for transform coefficient levels are encoded using binary probability modeling. The non-binary syntax element coefficient_absolute_level_less_two, for example, is converted to binary first, that is, mapped onto a sequence of binary (binary) decisions, and these binaries are encoded sequentially. The binary syntax element signifier_significant_coefficient, last_significant_coefficient_coefficient, major_one_absolute_coefficient, and coefficient_signal_ indicator is directly encoded. Each encoded binary (which includes the binary syntax elements) is associated with a context. A context represents a probability model for a class of encoded binaries. A measure of the probability for one of the two possible binary values is estimated for each context based on the values of the binaries that have already been encoded with the corresponding context. For various binaries related to transform encoding, the context used for encoding is selected based on already transmitted syntax elements or based on position within a block.

After encoding the significance map, the blocks are processed in reverse scanning order. As mentioned above, another concept is used in the first H.265 / HEVC test model. Transform blocks larger than 4x4 are partitioned into 4x4 blocks and the resulting 4x4 blocks are processed in scanning order, while the coefficients of 4x4 blocks are encoded in reverse scanning order. The following description is valid for all 4x4 blocks in the first test model of H.265 / HEVC and in H.264 / AVC and also for 8x8 blocks in H.264 / AVC and this description can also be applied to construction syntax element stream 138 and 226, respectively.

If a significant scan position, that is, the coefficient is non-zero, the binary syntax element greater_one_coefficient is transmitted within stream 138. Initially (within a block), the second context model is selected from the corresponding context for the syntax element coefficient_absolute_major_one. If the encoded value of any major_absolute_coefficient syntax element within the block equals one (that is, the absolute coefficient is greater than 2), context modeling switches back to the first context model in the set and uses this context model until the end of the block. Otherwise (all encoded values of absolute_coefficient_major_one within the block are zero and the corresponding absolute coefficient levels are equal to one), the context model is chosen depending on the number of syntax elements coefficient_absolute_major_one that have already been processed in the reverse scan of the considered block. The selection of the context model for the syntax element coefficient_major_one_one can be summarized by the following equation, where the current context model index C, + 1, is selected based on the index of the previous context model C, and the value of the element of previously coded syntax coefficient_absolute_major_one, which is represented by bint in the equation. For the first syntax element coefficient_major_one_one within the block, the index of the context model is set equal to C1 = 1.

The second syntax element for encoding the absolute transform coefficient levels, absolute_coefficient_level_less_two is only encoded, when the syntax element greater_one_absolute_coefficient is equal to one. The non-binary syntax element coefficient_absolute_level_less_two is converted to binary in a sequence of binaries and for the first binary of this conversion to binary; a context model index is selected as described below. The remaining binaries of the binary conversion are encoded with fixed contexts. The context for the first binary of the binary conversion is selected as follows. For the first syntax element coefficient_absolute_level_less_two, the first context model is selected from the set of context models for the first binary of the syntax element coefficient_absolute_level_legs_2, the index of the corresponding context model is set equal to C1 = 0. In addition to the syntax element coefficient_absolute_level_less_two, context modeling switches to the next context model in the set, where the number of context models in the set is limited 5. The selection of the context model can be expressed by the following formula, where the Current context model index Cr + 1 is selected based on the previous context model index C.

Cr + l ( Cr) = min (Cr 1.4)

As mentioned above, for the first syntax element coefficient_absolute_remains_less_two within a block, the context model index is set equal to Cr = 0. Note that different sets of context models can be defined for the syntax elements coefficient_absolute_majority_one and coefficient_absolute_maintain_two_ . Note also that for the first H.265 / HEVC test model, transform blocks larger than 4x4 can be partitioned into 4x4 blocks. Partitioned 4x4 blocks can be processed in scanning order and for each partitioned 4x4 block, a context set can be deduced based on the number of coefficients greater than one in the previous 4x4 block. For the first 4x4 block of a larger 4x4 transform block and for source 4x4 transform blocks, a separate context set can be used.

That is, in whatever case context-based encoding is used in the following description for any of the source symbols in which coefficient_absolute_major_one and coefficient_absolute_remains_minus_two are decomposed according to the following embodiments, then this context deduction may be used by allocator 114 and 212 and the VLC encoder / decoder 102 and 202, for example.

To reduce complexity in terms of number of binaries processed by CABAC or PIPE compared to prior art techniques and also in terms of computation complexity or also to increase coding efficiency, the embodiment outlined below describes an approach for encoding absolute levels by using different variable length codes for different partitions 1401 to 1403 in image and video encoders and decoders. The embodiments described below can, however, be applied to any kind of absolute levels for image and video encoders, such as motion vector differences or adaptive loop filter coefficients. Although the encoding of transform coefficient levels is performed as described below, the encoding of the significance map may remain as in the first H.265 / HEVC test model or as described above, or the encoding of the map of significance can be performed as in H.264 / AVC or otherwise.

As described above, encoding of the absolute transform levels is performed on multiple 1401-3 partitions. The coding scheme is illustrated in Figure 1b with three 1401-3 partitions as an example. The delimitations 142 and 144 of the schema are variable resulting in variable partition sizes. The coding scheme is done as follows.

A first entropy code is used to encode the first component or source symbol, that is, the level of absolute transform coefficient (z) if equal is less than limit1, or limit1 if it is not equal. If the absolute transform coefficient level is greater than or equal to the limit1 delimitation of the first partition 1401, then the limit1 delimitation (142) of the first partition (1401) is subtracted from the absolute transform coefficient level and the resulting value z 'is encoded with a second entropy code. If the level of the remaining absolute transform coefficient z 'is greater than or equal to a limit2-limit1 bounding for the second partition 1402, then the limit2-bounding1 bounding of the second partition is again subtracted from the level of absolute transform coefficient z 'and the resulting value is encoded with a third entropy code. Generally speaking, when the bounding of a partition is reached, the entropy code for the next partition in the bounding is used to encode the resulting value of the absolute transform coefficient levels minus the bounding of the corresponding partition.

Entropy codes can be simple variable length codes such as string length codes or lookup tables (eg Huffman code) or more complex entropy codes that employ probability models such as CABAC or PIPE. The number of partitions and the delimitation of partitions can be variable or dependent on the actual syntax element. The partitioning concept shown in Fig. 1b has the following benefits. The absolute transform coefficients are used below as an example, but it should be understood that they can be replaced by any other syntax element. As an example, the probability distribution of the absolute transform coefficient levels can have approximately a geometric distribution. Therefore, entropy codes optimized for geometric distributions can be used to encode absolute transform coefficient levels. But such a model is not always locally optimal, even if context modeling and probability model selection are employed. For example, for a transform block, the local absolute transform coefficient levels follow a distribution that is not geometric at all if it contains the same number of low and medium range transform coefficient levels, as long as the model can hold true (with a certain precision) for a specific number of blocks in image or video due to the law of large numbers. In this case, the geometric distribution is not a suitable model. Also, when considering absolute transform coefficient levels with large values, the distribution of these large values is often uniform. The partition concept allows different probability models for different levels of the absolute transform coefficient. For lower absolute values, more complex entropy codes may be applied for higher efficiency, while for higher absolute levels, less complex entropy codes may be used. to reduce complexity.

As mentioned above, suitable entropy codes are used for different partitions. In a preferred embodiment of the invention, three types of entropy codes are employed. The first entropy code uses PIPE. However, it should be noted that, according to an alternative, an entropic encoding method such as CABAC or any other arithmetic encoder can be used as an alternative. That is, the first s1 symbols (see Fig. 2b) can be encoded through the PIPE encoding path. Subdivider 100 and recombiner 220 act accordingly.

For the second type, Golomb codes and some truncated variants that include subsets (eg, Golomb-Rice codes) can be used. That is, the second s2 symbols (see Fig. 2b) can be encoded through said VLC codes in the VLC 102/202 encoder / decoder. Subdivider 100 and recombiner 220 act accordingly.

Exponential-Golomb codes are used as the third type. That is, the third symbols s3 (see Fig. 2b) can be encoded through said VLC codes in the VLC 102/202 encoder / decoder. Subdivider 100 and recombiner 220 act accordingly. Different VLC codes and different combinations of VLC and PIPE or VLC and arithmetic are possible.

While the first code is more complex but produces better compression performance, the second entropic code represents a reasonable compromise between complexity and performance. The latest entropy codes, for example Exponential-Golomb codes, have very low complexity. The encoding of different partitions is described below.

In a partition such as partition 1401, such as the symbol s1, it is to be encoded entropy using an entropic encoder employing probability models such as the PIPE 104 encoder (again, CABAC or any arithmetic encoder can be used in an alternative not described further herein), subdivider 120 directs it towards PIPE encoder 104. First, non-binary evaluated absolute transform coefficient levels can be converted to binary in symbolizer 122 using a binary conversion method. Conversion to binary maps the evaluated non-binary absolute transform coefficient levels into a binary binary sequence. Each binary in the binary chain is encoded with a context chosen by an allocator 114. Context modeling can be done by the first binary and set for subsequent binaries in the binary sequence since absolute_level_level_2 coefficient can be used in H.264 / AVC or a different context modeling for each binary in the binary chain. Note that the conversion to binary can be a variable length code such as Golomb codes or Exponential-Golomb codes or other variable length codes.

Next, a partition such as partition 1402 or s2 symbols can be encoded with a Golomb code, in this case in VLC encoder 102 and respectively decoded in VLC decoder. Golomb codes are a set of entropy codes designed for a geometrically distributed source. If the order of the Golomb code is zero, the Golomb code is also known as a unary code. The unary code refers to the conversion to binary of coefficient_absolute_level_less_dos in H.264 / AVC. Golomb codes are constructed as follows. For a specific Golomb parameter k, the value n is divided by the Golomb parameter k using integer division and the remainder r is calculated .

n

k

P = L

r = n - pk

After deducing the parameters specified by the formulas above, the value n can be coded with two parts. The first part, also called the prefix part, is a unary code. The resulting value p + 1 specifies the number of ones and an ending zero or vice versa. The value of remainder, also named as part remainder and indicated by r, is represented by a truncated binary code. According to a specific embodiment, the Golomb-Rice codes are used for encoding source symbols such as the s2 source symbols, the Golomb-Rice codes being a subset of the Golomb codes. Also, when such entropy codes are used for partitions 1401-3 that contain a delimitation, such as partition 1402, the alphabet of the respective source symbols (such as the s2 source symbols) is limited and the Golomb-Rice code may be modified so that the coding efficiency can be improved. The Golomb-Rice code parameter can be fixed or variable. If the parameter is variable, the parameter can be estimated as part of the context modeling stage. For example, if a source symbol s2 enters decoder VLC 102, the latter can determine the Golomb-Rice code parameter from a context of s2. Golomb-Rice codes are Golomb codes with parameters to the power of two. So they are based on division and multiplication by two and therefore they can be efficiently implemented in a binary architecture with shift and add operations. The relationship between the Golomb-Rice parameter and the Golomb parameter is therefore kGOLOMB = 2kRICE. In the case of the Golomb-Rice code, the remainder part is exactly the binary representation of the remainder value. For the Golomb-Rice parameter of zero, the resulting code is identical to the unary code and does not have a remainder part. For the parameter equal to one, the remainder part consists of a binary with two input symbols that share the same unary prefix. Some example tables for selected Golomb-Rice parameters are illustrated below.

Given the partition range such as for example limit2-limit1 in case of partition 1402 and the Golomb-Rice code parameter, truncation can be performed as follows. The Golomb-Rice parameter describes the number of binaries required to represent the remainder part, and the two to the power of the parameter value describes the number of values, which can be represented with the same prefix. These values form a prefix group. For example, for parameter zero, a prefix can only represent a specific value, while for parameter three, eight input values share the same prefix, and therefore a prefix group contains eight values for parameter three. For a limited font alphabet and a given Golomb-Rice code, the last prefix binary can be left out of the values in the last prefix group resulting in a fixed length prefix code.

For example, the range can be nine and the Golomb-Rice parameter two. For this example case the number of values that can be represented with the same prefix is four. The maximum value is nine, which also indicates that the edge has been exceeded and the following entropy code for the next partition is to be used. In this example case, the values from 0 - 3 have the prefix 0, the values from 4 -7 have the prefix 10 and 8 - 9 the prefix 110. Because the values 8 - 9 form the last group of prefix, its left zero can be removed and the values 8 - 9 can be represented by 11. In other words, you can imagine a source symbol s2 entered in the VLC 102 encoder with the number of possible values of s2 being 9 (= limit2-limit1 1) and the Golomb-Rice parameter for that source symbol is two. Next, a respective Golomb-Rice codeword would be produced by the VLC encoder 102 for that source symbol, which has a prefix as just described. For the remainder part of the codeword, the truncated code can be deduced as follows by the VLC 102 encoder. Normally, the Golomb-Rice codes parameter indicates the number of binaries of the remainder part. In a truncated case, not all of the rest binaries need to be encoded. For a truncated case, all values with fixed prefix are counted (for example a prefix binary is extracted). Note that the counted value is always smaller than or equal to the maximum number of values for a prefix because the code is truncated. If truncation of the remainder part is possible, deduction of the truncated remainder part for the last prefix group can proceed in the following steps. First, the largest number l is deduced from the power of two smaller than or equal to the number found. Then, in the second stage, the smallest number h is deduced from the power of two greater than the number found. The first value l writes the number of values in the prefix group with a remainder of binary h . All remainders of these values start with 0 followed by the binary representation of the remainder limited to the number of values in the remainder group. For the remaining values of the last prefix group, now treated as a new prefix group, the same procedure is performed except for the resulting values formed by the first group of remainder, the remainder starts with a 1. This procedure is carried out until all remains are deducted. As an example, the range is 14 and the parameter is three. The first group of prefix contains values from 0 - 7 and the second group of prefix values from 8 - 13. The second group The prefix contains six values. The parameters are l = 2 and h = 3. Thus, the first four values of the prefix groups are represented by a remainder with three binaries (a zero on the left and the binary representation to distinguish the four values). For the last two values, the same procedure is performed again. The parameters are l = 1 and h = 2. The rest of the last two values can now be represented as 10 and 11. Another example to demonstrate the method is a Golomb-Rice parameter of four and a range of ten. For this example, the parameters are l = 3 and h = 4. With these parameters, the truncated remainder part for the first eight values is represented by four binaries. The remaining two values have the same remainder part as in the previous example. If the range is nine for the previous example, the parameter for the second series is l = 0 and h = 1. The remainder part of the only value that remains is 1.

The third type of entropy codes can be Exponential-Golomb codes. They can be used for equally probable distributions (eg with parameter zero) such as the s3 source symbols. That is, the VLC encoder / decoder pair may be responsible for its encoding. As mentioned above, higher absolute transform coefficient levels are often evenly distributed. More precisely, the zero-order Exponential-Golomb code can be used to encode the last partition 1403. The beginning and therefore the edge 144 of the previous partition 1402 can be variable. The position of edge 144 can be controlled by VLC 102-200 encoder / decoder depending on precisely encoded / decoded source symbols 106, 108 and / or 110 or syntax elements 138 (or 218, 204 and / or 208 or syntax elements 226).

In a preferred embodiment, the number of partitions is three as shown in Fig. 1b and the boundaries 142 and 144 can be variable. For the first partition 1401, PIPE encoding can be used as explained above. However, CABAC can also be used as an alternative. In that case, the PIPE encoder / decoder pair would be replaced by an arithmetic encoder / decoder pair. The truncated Golomb-Rice codes can be used for the second 1402 partition and the zero-order Exponential-Golomb code can be used for the last 1403 partition.

In another preferred embodiment, the number of partitions 1401-3 is three and the first delimitation 142 is fixed, while the second delimitation 144 is variable. For the first partition 1401, CABAC or PIPE is used. The truncated Golomb-Rice codes can be used for the second 1402 partition, and the zero-order Exponential-Golomb code can be used for the last 1403 partition.

In another preferred embodiment, the number of partitions is equal to two. The first partition 1401 could use CABAC or PIPE. The second partition could use Golomb-Rice tale codes like 1402.

In another preferred embodiment, the number of partitions is equal to three while both edges 142 and 144 are variable. For example, for the first partition 1401, CABAC or PIPE is used, while the second partition 1402 can use the truncated Golomb-Rice code and the third partition 1403 uses the zero-order Exponential-Golomb code.

In a preferred embodiment, the boundary 142 of the first partition 1401 using CABAC or PIPE, which employs adaptive probability models is two. In this preferred embodiment, context modeling for the first binary can be performed as described for absolute_coefficient_major_one as described above and context modeling for the second binary can be performed as described for absolute_coefficient_level_less_two in H.264 / AVC , as also described above. The last context determination would be determined by allocators 114 and 212, respectively. In another preferred embodiment, the delimitation 142 of the first partition 1401 that uses entropy coding that employs probability modeling (eg, PIPE or CABAC) is two. For this preferred embodiment, context modeling for both the first and the second binary can be performed as described for absolute_coefficient_major_one in H.264 / AVC as described above. The evaluation of the context set as described for coefficient_absolute_major_one can be done separately for the second binary.

In a preferred embodiment, the delimitation 142 for the first partition 1401 that uses entropic coding that employs probability modeling (eg CABAC or PIPE) can be equal to one. For the single binary (or alphabet symbol) in the binary chain of the respective source symbol, context modeling can be performed as described for coefficient_absolute_major_one in H.264 / AVC previously described.

In a preferred embodiment, the delimitation 142 of the first partition 1401 that uses entropic coding that employs probability modeling (eg CABAC or PIPE) can be equal to three. Context modeling of the first and second binaries of the respective source symbol binary chain can be performed as absolute_coefficient_major_one in H.264 / AVC. Context modeling for the third binary can be performed as absolute_level_level_less_two in H.264 / AVC. The evaluation of the context set as described for absolute_coefficient_major_one can be done separately for the second binary.

In a preferred embodiment, a set of truncated Golomb-Rice codes can be used as entropy codes of the second partition 1402. Delimitation 144 of the second partition that specifies the start of the third partition 1403 depending on the entropy code parameter can be variable. Also, in this preferred embodiment, the Golomb-Rice parameter can be limited by three and the parameter selection can be performed as the context modeling for absolute_level_level_least_two in H.264 / AVC. The limit-limit2 range can be variable and can depend on the Golomb-Rice parameter. If the parameter is zero, the range is 8. For parameter one, the range is 10. In case of parameter two, the range is 12 and for parameter three, the range is equal to 16. In this preferred embodiment, the Golomb-Rice parameter is set to zero at the beginning of a transform coefficient block. For each level of code transform coefficient in the block greater than or equal to the first delimitation, the corresponding Golomb-Rice code is used. After level encoding (or decoding), the following evaluation is performed to update the Golomb-Rice parameter for next level encoding (or decoding) greater than or equal to the first delimitation. Note that the Golomb-Rice parameter cannot be decreased by using this form of adaptation.

The parameter adaptation rule can be summarized as follows, where k 1 indicates the Golomb-Rice parameter to be used for encoding the next level value and valort indicates the value previously encoded with the corresponding Golomb-Rice parameter kt

0 goes lo rt and i0, 1] to kt < 1

1 goes lo rt e [2, 3] to kt <2

Kt ⁺¹ = <2 goes what rt e [4, 5] to kt <3 (QQ)

3 goes rt> 5 ak <4

k in case c o n tra r io

In a preferred embodiment, a set of truncated Golomb-Rice codes can be used as entropy codes of a second partition 1402. The delimitation 144 of the second partition 1402 that specifies the start of the third partition 1403 depending on the parameter of the entropy code can be variable. Also in this preferred embodiment, the Golomb-Rice parameter can be limited to three and the parameter selection can be done as context modeling for absolute_level_level_least_two in H.264 / AVC. The range can be variable and depends on the Golomb-Rice parameter. If the parameter is zero, the range is 8. For parameter one, the range is 10. In case of parameter two, the range is 12 and for parameter three, the range is equal to 16. In this preferred embodiment, the Golomb-Rice parameter is set to zero at the beginning of the block. The adaptation of the Golomb-Rice parameter is performed as described by the equation (QQ). Note that the parameter cannot be decreased by using this form of adaptation.

In another preferred embodiment, a truncated Golomb-Rice code set can be used as entropy codes of second partition 1402. Delimitation 144 of second partition 1402 specifying the start of third partition 1403 depending on the entropy code parameter can be fixed. Also in this preferred embodiment, the Golomb-Rice parameter can be limited to three and the parameter selection can be done as context modeling for absolute_level_level_least_two in H.264 / AVC. The range of the second partition 1402 can be set to 14. In this preferred embodiment, the Golomb-Rice parameter can be set to zero at the beginning of the block. The adaptation of the Golomb-Rice parameter is performed as described by the equation (QQ). Note that the parameter cannot be decreased by using this form of adaptation.

In another preferred embodiment, a set of truncated Golomb-Rice codes can be used as entropy codes of a second partition 1402. The delimitation 144 of the second partition 1402 that specifies the start of the third partition 1403 depending on the parameter of the entropy code can be variable. Also in this preferred embodiment, the Golomb-Rice parameter can be limited to three and the parameter selection can be done as context modeling for absolute_level_level_least_two in H.264 / AVC. The range can be variable and depends on the Golomb-Rice parameter. If the parameter can be zero, the range can be 8. For parameter one, the range can be 10. In case of parameter two, the range can be 12 and for parameter three, the range can be equal to 16. In this Preferred embodiment, the Golomb-Rice parameter can be set to zero at the beginning of the block. The adaptation of the Golomb-Rice parameter is performed as described by the equation (QQ). Note that the parameter cannot be decreased by using this form of adaptation. And also note that direct switching is possible, for example from zero to three. In this preferred embodiment, the prefix portion of the Golomb-Rice codes is encoded with entropy codes that employ probability models. Context modeling can be done as for absolute_level_level_least_two in H.264 / AVC.

In another preferred embodiment, a fixed Golomb-Rice parameter can be used to encode all levels of transform coefficient in the current transform block. In this embodiment, the best parameter from the previous block can be calculated and used for the current transform block. For this embodiment, the range can be set to 14.

In another preferred embodiment, a fixed Golomb-Rice parameter can be used to encode all levels of transform coefficient in the current transform block. In this embodiment, the best parameter from the previous block can be calculated and used for the current transform block. For this embodiment, the range can be variable as previously described.

In another preferred embodiment, it is evaluated whether the already encoded (or decoded) neighborhood of the current scan index contains higher absolute transform coefficient levels than the previous bounding. For this preferred embodiment, the best parameter can be deduced by using neighbors in a local causal template.

Thus, the aforementioned embodiments describe, inter alia, an entropic encoding apparatus comprising a decomposer 136 configured to convert a sequence 138 of syntax elements to a sequence 106 of source symbols 106 by individually decomposing at least a subset of the elements of syntax in a respective number n of source symbols if, with i = 1, ..., n, the respective number n of source symbols depending on which of a sequence of n partitions 1401-3 within which a range is subdivided of values of the respective syntax elements, a value z of the respective syntax elements falls, so that a sum of values of the respective number of source symbols if it produces zy, if n> 1, for all i = 1, .. ., n-1, the value of si corresponds to a range of the i-th partition; a subdivider 100 configured to subdivide the source symbol sequence 106 into a first source symbol subsequence 108 and a second source symbol subsequence 110 such that all source symbols sx with x being a member of a first subset of {1. n} are contained within the first subsequence 108 and all source symbols s and being and a member of the second subset of {1 ... n} that is disjoint from the first subset, are contained within the second subsequence 110; a VLC encoder 102 configured to encode in a symbolic manner the source symbols of the first subsequence 108, and a PIPE or arithmetic encoder 104 configured to encode the second subsequence 110 of source symbols.

The z values of the subgroup of syntax elements can be absolute values. The second subset may be {1} with the sequence of n partitions arranged so that a partition of order p covers higher values in the range of values than a partition of order q for all p, qe {1 ... n} being p> q. n can be 3. The first subset can be {2, 3} configured the VLC encoder (102) to use a Golomb-Rice code to encode in a symbol level manner the source symbols s2, and an Exp-Golomb code to encode in a symbol level manner the s3 source symbols. More generally, 2 may be elements of the first subset with a VLC encoder (102) configured to use a Golomb-Rice code to encode the s2 source symbols in a symbol level manner and adapt a Golomb-Rice parameter, i.e. k, of the Golomb-Rice code according to the previously coded source symbols. The decomposer can be configured to adapt one or more of the boundaries between the partitions according to the previously encoded source symbols. Both adaptations can be combined. That is, the positions of the limits that limit the second partition can be adapted so that they are separated from each other so that the length of the Golomb-Rice code, that is, the number of code words in it, corresponds to (or prescribes) the length of the width of the second partition. The boundary for the separation from the first to the second partition can be defined differently as set by definition or adapted according to another context dependency, in which case the adaptation of k can define the position of the boundary separation of the second and third partition through the length of the Golomb-Rice code and the width of the second partition, respectively. By coupling the adaptation of k so that the width of the second partition corresponds to the length of the Golomb-Rice code, the efficiency of the code is used optimally. Adapting ka to the syntax element statistics allows adapting the width of the second partition so that the third partition can cover as much as possible so as to reduce the overall encoding complexity since less complex code may have been used for the third partition such as the Exp-Golomb code. Additionally, the length of the first partition can be restricted to j e {1, 2, 3} possible syntax element values, such as the three lowest levels. The syntax elements under consideration can be differentially encoded or represent a residual prediction as is the case with the transform coefficient levels exemplified above representing a residual prediction. The first source symbols s1 can be symbolized / de-symbolized by using a truncated unary code with the resulting binary js being encoded - partially or all of them - adaptively to the context or not, as mentioned above.

The subset of the syntax elements can encompass absolute transform coefficient levels of the absolute transform coefficients of image transform blocks with the absolute transform coefficient levels of a respective transform block that is arranged within the sequence ( 138) of syntax elements according to a sweep path that produces through the absolute transform coefficients of the respective transform blocks, in which the decomposer can be configured to adapt one or more boundaries between the partitions during the decomposition of the absolute transform coefficient levels of the absolute transform coefficients of a respective transform block depending on the already encoded absolute transform coefficient levels of the absolute transform coefficients of the respective transform blocks that precede in the scan order or depend on a position of the absolute transform coefficient level currently to decompose in the scan order, or be based on an evaluation of the already reconstructed absolute transform coefficient levels of coefficients transform values - both spatially and in scanning order - the position of the level of the absolute transform coefficient to be decomposed currently.

Additionally, the aforementioned embodiments describe among others an entropic decoding apparatus comprising a VLC decoder 200 configured to reconstruct in a codeword way source symbols of a first subsequence 204 of source symbols from codewords of a first stream bit 206; a PIPE or arithmetic decoder 202 configured to reconstruct a second subsequence 208 of source symbols; a composer 224 configured to compose a sequence 226 of syntax elements from the first source symbol subsequence 204 and the second source symbol subsequence 208 individually composing each syntax element from a respective number of source symbols, wherein the composer is configured to determine, in at least one subset of the syntax elements, the respective number n of source symbols if with i = 1 ... n depending on which of a sequence of n partitions 140 ⁱ -3 within which subdivides a range of values of the respective syntax elements, falls within a z-value of the respective syntax elements, by summing the values of the respective number of source symbols if from 1 to n as long as the value of if corresponds to a range in the i-th partition so that it gets the value of the z syntax element, in which the composer 224 is set to retrieve all the sx source symbols where x is a member of a first subset of {1 ... n} of the first subsequence (204) and all source symbols s and being y a member of a second subset of {1..n} that is disjoint with respect to the first subset, starting from the second subsequence 208. The z-values of the subgroup of the syntax elements can be absolute values. The second subset may be {1} with the sequence of n partitions arranged so that a partition of order p covers higher values in the range of values than a partition of order q for all p, qe {1 ... n} being p> q. n can be 3. The first subset can be {2, 3} with the VLC decoder 200 configured to use a Golomb-Rice code to reconstruct the s2 source symbols in code way, and an Exp-Golomb code to reconstruct in a code word way the s3 source symbols. More generally, 2 can be an element of the first subset with the VLC 102 decoder configured to use a Golomb-Rice code to reconstruct the s2 source symbols in code words and adapt a Golomb-Rice parameter of the Golomb-Rice code accordingly with previously reconstructed font symbols. The entropic decoding apparatus may further comprise a recombiner 220 configured to recombine the first source symbol subsequence 204 and the second source symbol subsequence to obtain source symbol sequence 218. Syntax elements can be of different types and the composer can be configured to perform individual composition depending on the type of syntax elements. The subset of the syntax elements can encompass absolute transform coefficient levels of absolute transform coefficients of the transform blocks of an image with the absolute transform coefficient levels of a respective transform block that is arranged within sequence 138 of syntax elements according to a scanning path leading through the absolute transform coefficients of the respective transform blocks, in which the composer can be configured to adapt one or more of the boundaries between the partitions during composition of the absolute transform coefficient levels of the absolute transform coefficients of a respective transform block depending on the already reconstructed absolute transform coefficient levels of absolute transform coefficients of the respective transform blocks that precede or depend on the scanning order a position The absolute transform coefficient level is currently to be composed in the scanning order, or based on an evaluation of the already reconstructed absolute transform coefficient levels of the neighboring transform coefficients - both spatially and in scanning order - the position of the level of absolute transform coefficient to be currently composed.

In relation to the embodiments combining PIPE coding with VLC coding using the decomposition according to Fig. 1b, the following is observed to repeat some aspects thereof in other words.

The mapping of a sequence of symbols into a bit stream and the reverse mapping have been described. Each symbol has a parameter (s) associated with it that are known simultaneously in encoder and decoder. The entropy codec contains multiple first-in-first-out (FIFO) buffers each assigned to subsets of parameter (s) that are associated with the symbols. For given parameter (s) of a symbol, the encoder assigns the symbol to the corresponding FIFO buffer. Since the encoder allocation rule is known on the decoder side, the decoder reads from the FIFO buffer that the encoder has assigned to the symbol.

Some syntax elements are encoded using standard variable length codes and written to a particular buffer. Other syntax elements are encoded using the concept of probability entropy of the partition interval (PIPE). In this, the symbols are converted to binary first and the resulting binaries are classified based on associated probability estimates. The probability estimate can be given or deduced from the measurement that can be performed simultaneously in encoder and decoder. A particular FIFO buffer contains symbols with estimated probability values that fall within the chosen subset of probabilities, so that entropic encoding can be improved. The improvement achieved by combining the PIPE concept with VLC is a reduction in complexity while still providing high coding efficiency. The symbols for which a standard VLC code is suitable are encoded with a simple VLC approach and low complexity, while other symbols for which the bit rate would be significantly increased by encoding with a VLC code are encoded with the more sophisticated PIPE concept.

Thus, to further reduce the complexity of entropic encoding, the symbols have been divided into two categories. Symbols in a first category can be represented well with VLC codes and do not require the most complex PIPE encoding, while symbols in a second category cannot be represented efficiently with VLC codes and a PIPE encoding for these symbols significantly reduces the required bit rate .

It should be appreciated that the above embodiments have been described with one VLC encoder and decoder only, but it is clear that this concept can be generalized to the use of two and more VLC encoders and decoders to be interleaved with PIPE encoders.

Although some aspects have been described in the context of an apparatus, it is evident that these aspects also represent a description of a corresponding method, where a block or device corresponds to a method stage or a characteristic of a method stage. Similarly, aspects described in the context of a method step also represent a description of a corresponding block or section or feature of a corresponding apparatus. Some or all of the steps of the method may be executed by (or using) a hardware apparatus, such as a microprocessor, a programmable computer, or an electronic circuit. In some embodiments, some or more of the most important method steps may be performed by such an apparatus.

The inventive encrypted / compressed signals can be stored on a digital storage medium or can be transmitted on a transmission medium such as a wireless transmission medium or a cable transmission medium such as the Internet.

Depending on certain implementation requirements, the embodiments of the invention can be implemented in hardware or software. Implementation can be done using a digital storage medium, for example a floppy disk, a DVD, a Blue-Ray, a CD, a ROM, a PROM, an EPROM, an EEPROM or a FLASH memory, which has readable control signals electronically stored therein, cooperating (or being able to cooperate) with a programmable computer system so that the respective method is performed. Therefore, the digital storage medium can be computer readable.

Some embodiments in accordance with the invention comprise a data carrier having electronically readable control signals, which is capable of cooperating with a programmable computer system, thereby performing one of the methods described herein.

Generally, the embodiments of the present invention can be implemented as a computer program product with a program code, the program code being operative for carrying out one of the methods when the computer program product is run on a computer. The program code can for example be stored on a machine-readable carrier.

Other embodiments comprise the computer program for performing one of the methods described herein, stored on a machine-readable carrier.

In other words, an embodiment of the inventive method is therefore a computer program having a program code for carrying out one of the methods described herein, when the computer program is run on a computer.

A further embodiment of the inventive methods is, therefore, a data carrier (or a digital storage medium, or a computer readable medium) comprising, recorded therein, the computer program for carrying out one of the methods. described in this document.

A further embodiment of the inventive method is, therefore, a data stream or a sequence of signals representing the computer program for carrying out one of the methods described herein. The data stream or signal sequence can for example be configured to be transferred over a data communication connection, for example via the Internet.

A further embodiment comprises a processing means, for example a computer, or a programmable logic device configured for or adapted to perform one of the methods described herein. A further embodiment comprises a computer having the computer program installed therein for carrying out one of the methods described herein.

In some embodiments, a programmable logic device (eg, a field of programmable gate arrays) may be used to perform some or all of the functionality of the methods described herein. In some embodiments, a field of programmable gate arrays can cooperate with a microprocessor to perform one of the methods described herein. Generally, the methods are preferably performed using a hardware apparatus.

The embodiments described above are merely illustrative of the principles of the present invention. It is understood that modifications and variations of the provisions and details described herein will be apparent to those skilled in the art. It is the intention, therefore, to be limited only by the scope of the impending patent claims and not by the specific details presented by way of description and explanation of the embodiments herein.

Claims (6)

1. Entropic coding apparatus comprising a decomposer (136) configured to convert a sequence (138) of syntax elements having a range of values that is subdivided into a sequence of N partitions (1401-3) into a sequence (106) of source symbols (106) by decomposing individually at least a subset of the syntax elements in a respective number n of source symbols if with i = 1 ... n, the respective number n of source symbols depending on which of the sequences of N partitions (14 O1-3 ) drops a value z of the respective syntax elements, so that a sum of values of the respective number of source symbols if it produces z, and, if n> 1, for all i = 1 ... n-1, the value whether it corresponds to a rank of the i-th partition; a subdivider (100) configured to subdivide the source symbol sequence (106) into a first source symbol subsequence (108) and a second source symbol subsequence (110) such that all source symbols sx, x being a member of a first subset of {1 ... N}, are contained within the first subsequence (108) and all source symbols sy, where y is a member of a second subset of {1 ... N} that is disjoint with respect to first subset, are contained within the second subsequence (110); a VLC encoder (102) configured to encode in a symbol level manner the source symbols of the first subsequence (108); and an arithmetic encoder (104) configured to encode the second subsequence (110) of source symbols, characterized in that the subgroup of the syntax elements encompasses levels of absolute transform coefficient of absolute transform coefficients of image transform blocks with levels absolute transform coefficient of a respective transform block representing a prediction residual and arranged within the sequence (138) of syntax elements according to a scanning path leading through the absolute transform coefficients of the respective transform blocks, where the decomposer is configured to adapt one or more of the boundaries between the divisions during the decomposition of the absolute transform coefficient levels of the absolute transform coefficients of a respective transform block depending on coefficient levels of absolute transform already encoded absolute transform coefficients of the respective preceding transform blocks in the scanning order or depending on a position of the absolute transform coefficient level to be currently decomposed in the scanning order, or based on an evaluation of the coefficient levels of Absolute transform already reconstructed from close transform coefficients, either spatially or in scanning order, to the position of the absolute transform coefficient level to decompose currently. 2. Entropic encoding apparatus according to claim 1, wherein the syntax element sequence (138) represents a video in a scalable manner. 3. Entropic coding apparatus comprising a VLC decoder (200) configured to reconstruct codeword level source symbols of a first source symbol subsequence (204) from codewords of a first bitstream (206); an arithmetic decoder (202) configured to reconstruct a second subsequence (208) of source symbols; a composer (224) configured to compose a sequence (226) of syntax elements with a value range that is subdivided into a sequence of N partitions (1401-3) from the first subsequence (204) of source symbols and the second subsequence (208) of source symbols by individually composing each syntax element from a respective number n of source symbols to, for at least a subgroup of the syntax elements, determining the respective number n of source symbols if with i = 1 ... n Depending on which of the sequence of N partitions (1401-3) in which a value range of the respective syntax elements is subdivided, a z value of the respective syntax elements falls, adding the values of the respective number of symbols source if from 1 to n as long as the value of si corresponds to an interval of the i-th partition to obtain the value of the z syntax element, where the composer (224) is configured to retrieve all s source symbols sx, where x is a member of a first subset of {1 ... N} from the first subsequence (204) and all source symbols sy, where y is a member of a second subset of {1 ... N} which is disjoint with respect to the first subset, from the second subsequence (208), characterized in that the subset of the syntax elements encompasses absolute transform coefficient levels of absolute transform coefficients of transform blocks of an image with the absolute transform coefficient levels of a respective transform block representing a prediction residual and is arranged within the sequence (138) of syntax elements according to a scanning path leading through the absolute transform coefficients of the respective transform blocks, where the composer is configured to adapt one or more of the boundaries between the divisions during the composition of the levels of the absolute transform coefficient of the absolute transform coefficients of a respective transform block depending on the already reconstructed absolute transform coefficient levels of the absolute transform coefficients of the respective preceding transform blocks in the order scanning or depending on a position of the absolute transform coefficient level to be currently composed in the scanning order, or based on an evaluation of the already reconstructed absolute transform coefficient levels of close transform coefficients, either spatially or in order scanning, to the position of the level of absolute transform coefficient to be currently composed. 4. Entropic decoding apparatus according to claim 3, wherein the sequence (226) of syntax elements represents a video in a scalable manner. 5. Entropic encoding method comprising converting a sequence (138) of syntax elements having a range of values that is subdivided into a sequence of N partitions (1401-3) into a sequence (106) of source symbols (106) by individually decomposing at least a subgroup of the syntax elements in a respective number n of source symbols if with i = 1 ... n, the respective number n of source symbols depending on which of the sequences of N partitions (1401-3) a z-value of the elements falls of respective syntaxes, so that a sum of values of the respective number of source symbols if produces z, and, if n> 1, for all i = 1 ... n-1, the value of si corresponds to a range of the i-th partition; subdivide the source symbol sequence (106) into a first source symbol subsequence (108) and a second source symbol subsequence (110) such that all source symbols sx, x being a member of a first subset of {1. ..N}, are contained within the first subsequence (108) and all source symbols sy, where y is a member of a second subset of {1 ... N} that is disjoint with respect to the first subset, are contained within the second subsequence (110); performing VLC encoding, encoding in a symbol level manner the source symbols of the first subsequence (108); and performing arithmetic encoding, encoding the second subsequence (110) of source symbols, characterized in that the subgroup of the syntax elements encompasses absolute transform coefficient levels of absolute transform coefficients of transform blocks of an image with the transform coefficient levels absolute of a respective transform block representing a prediction residual and arranged within the sequence (138) of syntax elements according to a scan path leading through the absolute transform coefficients of the respective transform blocks , where the composition comprises adapting one or more of the limits between the divisions during the composition of the levels of absolute transform coefficient of the absolute transform coefficients of a respective transform block depending on the levels of absolute transform coefficient already coded by coefficients d e absolute transform of the respective preceding transform blocks in the scanning order or depending on a position of the absolute transform coefficient level to be currently decomposed in the scanning order, or based on an evaluation of the absolute transform coefficient levels already reconstructed from close transform coefficients, either spatially or in scanning order, to the position of the absolute transform coefficient level to be currently composed. 6. Entropic decoding method comprising by VLC decoding, rebuilding source symbols at the codeword level of a first subsequence (204) of source symbols from codewords of a first bitstream (206); by arithmetic decoding, reconstruct a second subsequence (208) of source symbols; compose a sequence (226) of syntax elements with a value range that is subdivided into a sequence of N partitions (1401-3) from the first subsequence (204) of source symbols and the second subsequence (208) of source symbols composing individually each syntax element from a respective number n of source symbols to, for at least a subgroup of the syntax elements, determine the respective number n of source symbols if with i = 1 ... n depending on which of the sequence of N partitions (1401-3) in which a value range of the respective syntax elements is subdivided, a z value of the respective syntax elements falls, adding the values of the respective number of source symbols if from 1 to n always y when the value of si corresponds to an interval of the i-th partition to obtain the value of the z syntax element, where the composition (224) comprises recovering all the source symbols sx, x being a member of p first subset of {Í ... N} from the first subsequence (204) and all source symbols sy, being and member of a second subset of {1 ... N} that is disjoint from the first subset, from the second subsequence (208), characterized in that the subset of the syntax elements encompasses absolute transform coefficient levels of absolute transform coefficients of transform blocks of an image with the absolute transform coefficient levels of a respective transform block representing a prediction residual and is arranged within the sequence (138) of syntax elements according to a scanning path leading through the absolute transform coefficients of the respective transform blocks, where the composition comprises adapting one or more of the boundaries between the divisions during the composition of the levels of absolute transform coefficient of the absolute transform coefficients of a respective transform block depending on the already transformed absolute transform coefficient levels of absolute transform coefficients of the respective preceding transform blocks in the order of esc addendum or depending on a position of the absolute transform coefficient level to be currently decomposed in the scanning order, or based on an evaluation of the already reconstructed absolute transform coefficient levels of nearby transform coefficients, either spatially or in order of scan, at the position of the absolute transform coefficient level to be currently composed.